Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
From Lie Algebras to Quantum Groups
Helena Albuquerque Samuel Lopes Joana Teles Coimbra, Portugal, 28-30 June, 2006
28
ISBN: 978-989-95011-1-9 Título: From Lie Algebras to Quantum Groups Autor: Albuquerque, Helena; Lopes, Samuel; Teles, Joana Data: 08-05-2007 Editor: Centro Internacional de Matemática Morada: Almas de Freire, Coimbra Código Postal: 3040 COIMBRA Correio Electrónico: [email protected]: 239 802 370 Fax: 239 445 380
Contents
Foreword 1
Helena AlbuquerqueFrom Clifford algebras to Cayley algebras 3
Isabel Cunha, Alberto ElduqueFreudenthal’s magic supersquare in characteristic 3 17
G. LuztigOn graded Lie algebras and intersection cohomology 27
Carlos Moreno, Joana TelesOn the quantization of a class of non-degenerate triangular Lie bialgebras
over IK[[~]] 33
J. M. CasasOn the central extension of Leibniz n-algebras 57
Cristina Draper, Candido MartinFine gradings on some exceptional algebras 79
Vadim Kaimanovich, Pedro FreitasNotes on a general compactification of symmetric spaces 93
Darren Funk-NeubauerTridiagonal pairs, the q-tetrahedron algebra, Uq(sl2), and Uq(sl2) 99
Dimitri Gurevich, Pavel SaponovQuantum Lie algebras via modified reflection equation algebra 107
Stephane LaunoisOn the automorphism groups of q-enveloping algebras of nilpotent Lie algebras 125
Antonio Fernandez Lopez, Esther Garcıa, Miguel Gomez LozanoAn Artinian theory for Lie algebras 145
J. N. Alonso Alvarez, J. M. Fernandez Vilaboa, R. Gonzalez RodriguezQuantum grupoids with projection 151
Natasha RozhkovskayaShifted determinants over universal enveloping algebra 175
Erhard Neher, Maribel ToconGraded-simple Lie algebras of type B2 and Jordan systems covered by a triangle 187
Filippo VivianiRestricted simple Lie algebras and their infinitesimal deformations 197
Workshop From Lie Algebras to Quantum Groups
Helena Albuquerque∗ Samuel Lopes† Joana Teles‡
Foreword
This workshop brought together leading specialists in the topics of Lie algebras, quantum
groups and related areas. It aimed to present the latest developments in these areas as well
as to stimulate the interaction between young researchers and established specialists.
We remark on the significant role that, for the last decades, the theory of algebras has
played in the development of some areas of physics, and conversely, the importance that the
growth of physics has had in the implementation of new algebraic structures. In fact, with the
development of physics, more complex algebraic structures have arisen and the mathematical
structures that were used to explain certain physical phenomena became insufficient. For
example, there are two structures of considerable importance in contemporary mathematical
research that are deeply connected to the theory of Lie algebras: Lie superalgebras and
quantum groups.
The new supersymmetry theories that appeared in the 80’s presented, within the same
structure, particles that satisfied commutation relations and others that satisfied anti-commu-
tation relations. The algebras that existed up until then did not exhibit such a structure, and
that led to the emergence of a new mathematical structure: Lie superalgebras. The genesis of
quantum groups is quite similar. Introduced independently by Drinfel’d and Jimbo in 1985,
quantum groups appeared in connection with a quantum mechanical problem in statistical
mechanics: the quantum Yang-Baxter equation. It was realized then that quantum groups had
far-reaching applications in theoretical physics (e.g. conformal and quantum field theories),
knot theory, and virtually all other areas of mathematics. The notion of a quantum group is
also closely related to the study of integrable dynamical systems, from which the concept of
a Poisson-Lie group emerged, and to the classical Weyl-Moyal quantization.
∗Departamento de Matematica, Universidade de Coimbra, Portugal.†Departamento de Matematica Pura, Universidade do Porto, Portugal.‡Departamento de Matematica, Universidade de Coimbra, Portugal.
1
The Organizing Committee members were: Helena Albuquerque, Departamento de Matematica, Universidade de Coimbra, Portugal Samuel Lopes, Departamento de Matematica Pura, Universidade do Porto, Portugal Joana Teles, Departamento de Matematica, Universidade de Coimbra, Portugal
The Scientific Committee members were: Helena Albuquerque, Departamento de Matematica, Universidade de Coimbra, Portugal Georgia Benkart, Department of Mathematics, University of Wisconsin-Madison, USA Alberto Elduque, Departamento de Matematicas, Universidad de Zaragoza, Spain George Lusztig, Department of Mathematics, Massachusetts Institute of Technology,
USA Shahn Majid, School of Mathematical Sciences, Queen Mary, University of London,
UK Carlos Moreno, Departamento de Fisica Teorica de la Universidad Complutense de
Madrid, Spain Michael Semenov-Tian-Shansky, Departement de Mathematiques de l’Universite de Bour-
gogne, France
The Editors wish to thank the following Institutions for their financial support: Fundacao para a Ciencia e a Tecnologia Centro de Matematica da Universidade de Coimbra Centro de Matematica da Universidade do Porto Centro Internacional de Matematica
2
From Clifford algebras to Cayley algebras
Helena Albuquerque∗
1 Introduction
In this paper I present some of the main results on the theory of quasiassociative algebras
that were published untill June 2006, in a study developed in collaboration with Shahn Majid,
Alberto Elduque, Jose Perez-Izquierdo and A.P. Santana. It is a kind of structure that has
begin to be studied by the author and S. Majid in 1999 in the context of the theory of
cathegories of quasi-Hopf algebras and comodules [6] that generalizes the known theory of
associative algebras. Octonions appears in [6] as an algebra in the monoidal cathegory of
G-graded vector spaces. At the algebraic point of view this theory presents new examples of
a kind of quasilinear algebra that deals with a set of matrices endowed with a non associative
multiplication. In [6] the notion of a dual quasi-Hopf algebra was obtained dualizing Drinfeld’s
axioms.
A dual quasibialgebra is a (H,∆, ǫ, φ) where the coproduct ∆ : H → H ⊗ H and counit
ǫ : H → k form a coalgebra and are multiplicative with respect to a ‘product’ H ⊗H → H.
Besides, H is associative up to‘conjugation’ by φ in the sense
∑
a(1) · (b(1) · c(1))φ(a(2), b(2), c(2)) =∑
φ(a(1), b(1), c(1))(a(2) · b(2)) · c(2) (1)
for all a, b, c ∈ H where ∆h =∑h(1) ⊗ h(2) is a notation and φ is a unital 3-cocycle in the
sense∑φ(b(1), c(1), d(1))φ(a(1), b(2)c(2), d(2))φ(a(2), b(3), c(3)) =
∑φ(a(1), b(1), c(1)d(1))φ(a(2)b(2), c(2), d(2)),
(2)
for all a, b, c, d ∈ H, and φ(a, 0, b) = ǫ(a)ǫ(b) for all a, b ∈ H. Also φ is convolution-invertible
in the algebra of maps H⊗3 → k, i.e. that there exists φ−1 : H⊗3 → k such that
∑φ(a(1), b(1), c(1))φ
−1(a(2), b(2), c(2)) = ǫ(a)ǫ(b)ǫ(c)
=∑φ−1(a(1), b(1), c(1))φ(a(2), b(2), c(2))
(3)
∗Departamento de Matematica, Universidade de Coimbra, Apartado 3008, 3001-454 Coimbra, Portugal.
E-mail:[email protected]. Supported by CMUC-FCT.
3
for all a, b, c ∈ H. A dual quasibialgebra is a dual quasi-Hopf algebra if there is a linear map
S : H → H and linear functionals α, β : H → k such that
∑(Sa(1))a(3)α(a(2)) = 1α(a),
∑a(1)Sa(3)β(a(2)) = 1β(a),
∑φ(a(1), Sa(3), a(5))β(a(2))α(a(4)) = ǫ(a),
∑φ−1(Sa(1), a(3), Sa(5))α(a(2)β(a(4)) = ǫ(a)
(4)
for all a ∈ H.
H is called dual quasitriangular if there is a convolution-invertible map R : H ⊗H → k
such that
R(a · b, c) =∑
φ(c(1), a(1), b(1))R(a(2), c(2))φ−1(a(3), c(3), b(2))R(b(3), c(4))φ(a(4), b(4), c(5)),
(5)
R(a, b · c) =∑φ−1(b(1), c(1), a(1))R(a(2), c(2))φ(b(2), a(3), c(3))R(a(4), b(3))φ
−1(a(5), b(4), c(4))(6)
∑
b(1) · a(1)R(a(2), b(2)) =∑
R(a(1), b(1))a(2) · b(2) (7)
for all a, b, c ∈ H.
A corepresentation or comodule under a coalgebra means vector space V and a map
β : V → V ⊗H obeying (id⊗ ∆) β = (β ⊗ id) ∆ and (id ⊗ ǫ) β = id.
If F is any convolution-invertible map, F : H ⊗H → k obeying F (a, 0) = F (0, a) = ǫ(a)
for all a ∈ H (a 2-cochain) and H is a dual quasi-Hopf algebra then so is HF with the new
product .F , φF , RF , αF , βF given by
a ·F b =∑F−1(a(1), b(1))a(2)b(2)F (a(3), b(3))
φF (a, b, c) =∑F−1(b(1), c(1))F
−1(a(1), b(2)c(2))φ(a(2), b(3), c(3))F (a(3)b(4), c(4))F (a(4), b(5))
αF (a) =∑F (Sa(1), a(3))α(a(2))
βF (a) =∑F−1(a(1), Sa(3))β(a(2))
RF (a, b) =∑F−1(b(1), a(1))R(a(2), b(2))F (a(3), b(3))
(8)
for all a, b, c ∈ H. This is the dual version of the twisting operation or ‘gauge equivalence’
of Drinfeld, so called because it does not change the category of comodules up to monoidal
equivalence.
So the notion of comodule quasialgebra with the approach described in [6] is,
Definition 1.1. A G-graded algebra A is quasiassociative if is a G-graded vector space A =
⊕g∈GAg that satisfies
(ab)c = φ(a, b, c)a(bc),∀a∈Aa ,b∈Ab,c∈Ac, (9)
for any invertible group cocycle φ : G×G×G→ K∗ with
φ(x, y, z)φ(y, z, t) =φ(xy, z, t)φ(x, y, zt)
φ(x, yz, t), φ(x, 0, y) = 1,∀x, y, z, t ∈ G. (10)
4
(⇒ A0 is an associative algebra and Ag is an A0- bimodule, ∀g ∈ G)
Examples of quasiassociative algebras are KFG algebras: KFG is the same vector space
as the group algebra KG but with a different product a.b = F (a, b)ab,∀a,b∈G, where F is a
2-cochain on G. For this class of algebras the cocycle φ depends on F ,
φ(x, y, z) =F (x, y)F (xy, z)
F (y, z)F (x, yz), x, y, z ∈ G (11)
and measures the associativity of the algebra. Also the map R(x, y) = F (x,y)
F (y,x), x, y ∈ G has an
important role in the study of these algebras because it measures the commutativity.
2 Examples of KFG algebras
In this section we exemplify some results that we have proved in [4,7]. Some known classes of
KFG algebras were studied and were characterized by the properties of the cochain F . For
example, we have studied some conexions between alternative KFG algebras and composition
algebras,
Theorem 2.1. kFG is an alternative algebra if and only if for all x, y ∈ G we have,
φ−1(y, x, z) +R(x, y)φ−1(x, y, z) = 1 +R(x, y)
φ(x, y, z) +R(z, y)φ(x, z, y) = 1 +R(z, y).(12)
In this case, φ(x, x, y) = φ(x, y, y) = φ(x, y, x) = 1
Theorem 2.2. If G ≃ (Z2)n then the Euclidean norm quadratic function defined by q(x) = 1
for all x ∈ G makes kFG a composition algebra if and only if F 2(x, y) = 1 for all x, y ∈ G
and F (x, xz)F (y, yz) + F (x, yz)F (y, xz) = 0 for all x, y, z ∈ G with x 6= y.
Theorem 2.3. If σ(x) = F (x, x)x for all x ∈ G is a strong involution, and F 2 = 1, then the
following are equivalent,
i) kFG is an alternative algebra,
ii) kFG is a composition algebra.
Cayley Dickson Process was studied also in [6], and it was proved that after applying
this process to a KFG algebra we obtain another KFG algebra related to the first one which
properties are predictable. In fact,
Theorem 2.4. Let G be a finite abelian group, F a cochain on it (kFG is a G-graded quasial-
gebra). For any s : G→ k∗ with s(e) = 1 we define G = G×Z2 and on it the cochain F and
function s,
5
F (x, y) = F (x, y), F (x, vy) = s(x)F (x, y),
F (vx, y) = F (y, x), F (vx, vy) = αs(x)F (y, x),
s(x) = s(x), s(vx) = −1 for all x, y ∈ G.
Here x ≡ (x, e) and vx ≡ (x, ν) denote elements of G, where Z2 = e, ν with product
ν2 = e.
If σ(x) = s(x)x is a strong involution, then kFG is the algebra obtained from Cayley-
Dickson process applied to kFG.
We conclude that if G = (Z2)n and F = (−1)f then the standard Cayley-Dickson process
has G = (Z2)n+1 and F = (−1)f . Using the notation ~x = (x1, · · · , xn) ∈ (Z2)
n where
xi ∈ 0, 1 we have
f((~x, xn+1), (~y, yn+1)) = f(~x, ~y)(1 − xn+1) + f(~y, ~x)xn+1 + yn+1f(~x, ~x) + xn+1yn+1.
Theorem 2.5.
(i) The ‘complex number’ algebra has this form with G = Z2, f(x, y) = xy, x, y ∈ Z2 where
we identify G as the additive group Z2 but also make use of its product.
(ii) The quaternion algebra is of this form with
G = Z2 × Z2, f(~x, ~y) = x1y1 + (x1 + x2)y2
where ~x = (x1, x2) ∈ G is a vector notation.
(iii) The octonion algebra is of this form with
¯G = Z2 × Z2 × Z2,¯f(~x, ~y) =
∑
i≤j
xiyj + y1x2x3 + x1y2x3 + x1x2y3.
(iv) The 16-onion algebra is of this form with
¯G = Z2 × Z2 × Z2 × Z2
and
¯f(~x, ~y) =
∑
i≤j
xiyj +∑
i6=j 6=k 6=i
xixjyk +∑
distinct i,j,k,l
xixjykyl +∑
i6=j 6=k 6=i
xiyjykx4.
Now consider a n dimensional vector space V over a field K with characteristic not 2.
Define in V a nondegenerate quadratic form q . We know that there is an orthogonal basis
e1, · · · , en of V with q(ei) = qi for some qi 6= 0. The Clifford algebra C(V,q), is the
associative algebra generated by 1 and ei with the relations e2i
= qi.1, eiej +ejei = 0,∀i 6= j.
The dimension of C(V,q) is 2n and it has a canonical basis ei1 · · · eip | 1 ≤ i1 < i2 · · · < ip ≤
n.
In [9] we have studied Clifford Algebras as quasialgebras KFG and some of their repre-
sentations,
6
Theorem 2.6. The algebra kFZn
2can be identified with C(V,q), where F ∈ Z2(G, k) is
defined by F (x, y) = (−1)
∑
j<i
xiyj n∏
i=1
qxiyi
iwhere x = (x1, · · · xn) ∈ Zn
2and twists kG into a
cotriangular Hopf algebra with R(x, y) = (−1)ρ(x)ρ(y)+x·y where ρ(x) =∑
ixi and x · y is the
dot product of Z2-valued vectors.
C(V,q) is a superalgebra considering the Z2 graduation induced by the map σ(ex) =
(−1)ρ(x)ex extended linearly.
Theorem 2.7. C(V ⊕W,q ⊕ p) ≃ C(V,q) ⊗ C(W,p) as super algebras.
(This result is known in the classical theory but with our approach is easier to prove. It
is clear from the form of F that F ((x, x′), (y, y′)) = F (x, y)F (x′, y′)(−1)ρ(x′)ρ(y) where ex
is a basis of V and ex′ of W with the multiplication defined in the algebra product by
(a⊗ c)(b⊗ d) = a · b⊗ c · d(−1)ρ(c)ρ(b).)
We now use the above convenient description of Clifford algebras to express a ‘doubling
process’ similar to the Cayley Dickson process. Let A be a finite-dimensional algebra with
identity 1 and σ an involutive automorphism of A. For any fixed element q ∈ k∗ there is a
new algebra of twice the dimension, A = A⊕Av, which multiplication is given by
(a+ bv) · (c+ dv) = a · c+ qb · σ(d) + (a · d+ b · σ(c))v,
and with a new involutive automorphism,
σ(a+ vb) = σ(a) − σ(b)v.
We will say that A is obtained from A by Clifford process.
Theorem 2.8. Let G be a finite Abelian group and F a cochain as above. So kFG is a G-
graded quasialgebra and for any s : G→ k∗ with s(e) = 1 and any q ∈ k∗, define G = G×Z2
and
F (x, yv) = F (x, y) = F (x, y),
F (xv, y) = s(y)F (x, y), F (xv, yv) = qs(y)F (x, y),
s(x) = s(x), s(xv) = −s(x) for all x, y ∈ G.
Here x ≡ (x, e) and xv ≡ (x, η) where η with η2 = e is the generator of the Z2. If σ(ex) =
s(x)ex is an involutive automorphism then kFG is the Clifford process applied to kFG.
7
Theorem 2.9. For any s : G → k∗ and q ∈ k∗ as above the kFG given by the generalised
Clifford process has associator and braiding
φ(x, yv, z) = φ(x, y, zv) = φ(x, yv, zv) = φ(x, y, z),
φ(xv, y, z) = φ(xv, yv, z) = φ(xv, y, zv) = φ(xv, yv, zv) = φ(x, y, z)s(yz)
s(y)s(z)
R(x, y) = R(x, y), R(xv, y) = s(y)R(x, y)
R(x, yv) =R(x, y)
s(x),
R(xv, yv) = R(x, y)s(y)
s(x).
Theorem 2.10. If s defines an involutive automorphism σ then
1. kFG is alternative iff kFG is alternative and for all x, y, z ∈ G, either φ(x, y, z) = 1 or
s(x) = s(y) = s(z) = 1.
2. kFG is associative iff kFG is associative.
If F and s have the form F (x, y) = (−1)f(x,y), s(x) = (−1)ξ(x), q = (−1)ζ for some Z2-
valued functions f, ξ and ζ ∈ Z2 then the generalised Clifford process yields the same form with
G = Zn+1
2and f((x, xn+1), (y, yn+1)) = f(x, y)+(yn+1ζ+ξ(y))xn+1, ξ(x, xn+1) = ξ(x)+xn+1.
Theorem 2.11. Starting with C(r, s) the Clifford process with q = 1 yields C(r+1, s). With
q = −1 it gives C(r, s + 1). Hence any C(m,n) with m ≥ r, n ≥ s can be obtained from
successive applications of the Clifford process from C(r, s).
So as an immediate consequence of the last theorem we can say that starting with C(0, 0) =
k and iterating the Clifford process with a choice of qi = (−1)ζi at each step, we arrive at the
standard C(V,q) and the standard automorphism σ(ex) = (−1)ρ(x)ex.
Besides we must note that the Clifford Process is a twisting tensor product A = A ⊗σ
C(k, q) where C(k, q) = k[v] with the relation v2 = q, and va = σ(a)v for all a ∈ A.
3 Quasirepresentations and quasilinear algebra
In [9] we have extended Clifford process also to representations proving that,
8
Theorem 3.1. If W is an irreducible representation of A not isomorphic to Wσ defined by
the action of σ(a) then W = W ⊕Wσ is an irreducible representation π of A obtained via the
Clifford process with q. Here
π(v) =
(
0 1
q 0
)
,
π(a) =
(
a 0
0 σ(a)
)
are the action on W ⊕W in block form (here π(a) is the explicit action of a in the direct sum
representation W ⊕Wσ). If W,Wσ are isomorphic then W itself is an irreducible representa-
tion of A for a suitable value of q.
More generaly we have defined an action of a quasialgebra in [6],
Theorem 3.2. A representation or ‘action’ of a G-graded quasialgebra A is a G-graded vector
space V and a degree-preserving map : A⊗ V → V such that (ab) v = φ(|a|, |b|, |v|)a (b
v), 1 v = v on elements of homogeneous degree. Here |a v| = |a||v|.
Theorem 3.3. Let |i| ∈ G for i = 1, · · · , n be a choice of grading function. Then the usual
n× n matrices Mn with the new product
(α · β)ij =∑
k
αi
kβkj
φ(|i|, |k|−1, |k||j|−1)
φ(|k|−1, |k|, |j|−1),∀α, β ∈Mn
form a G-graded quasialgebra Mn,φ, where |Eij| = |i||j|−1 ∈ G is the degree of the usual basis
element of Mn. An action of a G-graded quasialgebra A in the n-dimensional vector space
with grading |i| is equivalent to an algebra map ρ : A→Mn,φ.
Several properties of quasilinear algebra can be proved using this class of matrices with
a “quasi-associative” multiplication. For example we can study a quasi-LU decomposition
for some matrix X ∈ Mn,φ:(Note that this study generalizes the known LU decomposition
of a matrix in the usual associative algebra of square matrices n × n.) We began studying
elementary “quasi-matrices” and “quasi-elementary” operations and we can conclude that,
Theorem 3.4. Let X ∈ Mn,φ. There are quasi-permutation matrices Pi1,j1, Pi2,j2
, ..., Pik ,jk,
an upper triangular matrix U and a lower triangular matrix L with lii = φ(i−1, i, i−1) such
that Pi1,j1(Pi2,j2
(...(Pik ,jkX))...) = L.U
9
4 Division quasialgebras
In [7] we have studied cocycles and some properties of quasialgebras graded over the ciclic
group Zn. Using the definition of a 3-cocycle in a group G we proved that a Z2-graded
quasialgebra is either an associative superalgebra or an antiassociative superalgebra (aaq-
algebra) with cocycle defined by,φ(x, y, z) = (−1)xyz,∀x, y, z,∈ Z2.
In Z3 every cocycle has the form
φ111 = α, φ112 = β, φ121 =1
ωα, φ122 =
ω
β, φ211 =
α
βω,
φ212 = αω, φ221 =β
ωα, φ222 =
ω
α
for some α, β ∈ K∗ with ω a cubic root of the unity. Here φ(1, 1, 1) = φ111, etc. is a shorthand.
Definition 4.1. The quasiassociative algebra A = ⊕g∈GAg is said to be a quasiassociative
division algebra if it is unital (1 ∈ A0) and any nonzero homogeneous element has a right and
a left inverse. Given such an algebra we denote the right inverse of 0 6= u ∈ Ag by u−1 and
the left inverse by u−1
L. Notice that, since A0 is a division associative algebra, the left and
right inverses of any nonzero element of A0 coincide.
For any nonzero u ∈ Ag, the elements u−1 and u−1
Lare in A−g, and we have u−1 =
φ(−g, g,−g)u−1
L. For 0 6= u ∈ Ag and 0 6= w ∈ Ah, we have (uw)−1 = φ(h,−h,−g)
φ(g,h,−g−h)w−1u−1.
It is easy to prove that for division quasiassociative algebra, the null part A0 is a division
associative algebra and Ag is an A0-bimodule satisfying Ag = A0ug = ugA0, for any nonzero
ug ∈ Ag, any g ∈ G.
In [5] we have studied division antiassociative algebras and characterized antiassociative
algebras with semisimple (artinian) even part and odd part that is a unital bimodule for the
even part.
Theorem 4.1. Given a division (associative) algebra D, σ an automorphism of D such that
there is an element a ∈ D∗ with σ2 = τa : d 7→ ada−1 and with σ(a) = −a, on the direct sum
of two copies of D: ∆ = D ⊕Du (here u is just a marking device), define the multiplication
(d0 + d1u)(e0 + e1u) = (d0e0 + d1σ(e1)a) + (d0e1 + d1σ(e0))u. (13)
Then with ∆0 = D and ∆1 = Du, this is easily seen to be a division aaq-algebra. We
will denote by < D,σ, a > the division aaq-algebra ∆ described before. The aaq-algebras
< D,σ, a > exhaust, up to isomorphism, the division aaq-algebras with nonzero odd part.
10
In general, it was proved in [2] that,
Theorem 4.2. Let D be a division associative algebra over K, G a finite abelian group and
φ : G × G × G → K∗ a cocycle. Suppose that for each g, h, l ∈ G, there are automorphisms
ψg of D and non zero elements cg,h of D satisfying
ψgψh = cg,hψg+hc−1
g,h(14)
and
cg,hcg+h,l = φ(g, h, l)ψg(ch,l)cg,h+l (15)
In the direct sum of |G| copies of D, ∆ = ⊕g∈GDug (ug is a marking device with u0 = 1),
consider the multiplication defined by
d1(d2ug) = (d1d2)ug
(d1ug)d2 = (d1ψg(d2))ug
(d1ug)(d2uh) = (d1ψg(d2)cg,h)ug+h,
(16)
for d1, d2 ∈ D and g, h ∈ G. Then with ∆0 = D and ∆g = Dug,∆ is a quasiassociative
G-graded division algebra. Conversely, every quasiassociative G graded division algebra can
be obtained this way.
As an interesting particular case of division quasialgebras we have studied the alternative
ones in [4]. We called strictly alternative algebras the alternative algebras that are not
associative.
Theorem 4.3. There are no strictly alternative division quasialgebras over fields of charac-
teristic 2.
Let A = ⊕g∈GAg be a strictly alternative division quasialgebra. Then G/N ∼= Z2×Z2×Z2
where N = x ∈ A : (x,A,A) = 0.
In [4] we have proved that any strictly alternative division quasialgebra over a field of
characteristic different from 2 can be obtained by a sort of “graded Cayley-Dickson doubling
process” built from the associative division algebra AN = ⊕g∈NAg.
Definition 4.2. Let K be a field, G an abelian group and N ≤ S ≤ T ≤ G a chain of
subgroups with [T : S] = 2 and T = S ∪ Sg with g2 ∈ N . Let β : G 7→ K be the map given by
β(g) = 1 if g ∈ N and β(g) = −1 if g 6∈ N . Let F : S × S → K× be a 2-cochain such that
F (s1, s2)
F (s2, s1)= β(s1)β(s2)β(s1s2)
(= (∂β)(s1, s2)
)
11
for any s1, s2 ∈ S and let 0 6= α ∈ K. Then the 2-cochain F = F(T,g,α) : T × T → K× defined
by
F (x, y) = F (x, y)
F (x, yg) = F (y, x)
F (xg, y) = β(y)F (x, y)
F (xg, yg) = β(y)F (y, x)α
(17)
for any x, y ∈ S, is said to be the 2-cochain extending F by means of (T, g, α).
Theorem 4.4. Let k be a field of characteristic 6= 2 and K/k a field extension. Let G be an
abelian group, N a subgroup of G such that G/N ∼= Z2 × Z2 × Z2 and N1 and N2 subgroups
of G such that N < N1 < N2 < G. Let β : G 7→ K× be the map given by β(x) = 1
for x ∈ N and β(x) = −1 for x 6∈ N and F0 : N ×N 7→ K× a symmetric 2-cocycle. Let
g1, g2, g3 ∈ G with N1 = 〈N, g1〉, N2 = 〈N1, g2〉 and G = 〈N2, g3〉, and let α1, α2, α3 be nonzero
elements in K. Consider the extended 2-cochains F1 = (F0)(N1,g1,α1), F2 = (F1)(N2,g2,α2) and
F3 = (F2)(G,g3,α3). Then KF3
G is a strictly alternative division quasialgebra over k.
Conversely, if k is a field and A is a strictly alternative division quasialgebra over k then
the characteristic of k is 6= 2 and there are K,G,N,F0, α1, α2, α3 satisfying the preceding
conditions such that A ∼= KF3G.
5 Quasialgebras with simple null part
In [5] we have classified antiassociative quasialgebras which even part is semisimple (artinian)
and the odd part is a unital bimodule for the even part. We have concluded that,
Theorem 5.1. Any unital aaq-algebra with semisimple even part is a direct sum of ideals
which are of one of the following types:
(i) A = A0 ⊕A1 with A0 simple artinian and (A1)2 6= 0.
(ii) A = A0 ⊕ A1 with A0 = B1 ⊕ B2, where B1 and B2 are simple artinian ideals of A0,
and A1 = V12 ⊕ V21, where V 2
12= 0 = V 2
21, V12V21 = B1 and V21V12 = B2.
(iii) A trivial unital aaq-algebra with a semisimple even part.
To better understand the last theorem consider two types of aaq-algebras with semisimple
artinian even part: Let ∆ be a division aaq-algebra and consider a natural number n. The set Matn(∆) is
an aaq-algebra whose even part is Matn(∆0);
12
Let ∆ be a division aaq-algebra and consider natural numbers n,m. The set Matn,m(D)
of (n+m) × (n+m) matrices over D, with the chess-board Z2-grading:
Matn,m(D)0 =
(
a 0
0 b
)
: a ∈Matn(D), b ∈Matm(D)
Matn,m(D)1 =
(
0 v
w 0
)
: v ∈Matn×m(D), w ∈Matm×n(D)
(18)
with multiplication given by,
(
a1 v1
w1 b1
)
·
(
a2 v2
w2 b2
)
=
(
a1a2 + v1w2 a1v2 + v1b2
w1a2 + b1w2 −w1v2 + b1b2
)
.
Matn,m(D) is an aaq-algebra, whose even part is isomorphic to Matn(D)×Matm(D).
Then,
Theorem 5.2. Any unital aaq-algebras with semisimple even part A = A0 ⊕ A1 is a finite
direct sum of ideals A = A1 ⊕ · · · ⊕Ar ⊕ A1 ⊕ · · · ⊕ As ⊕ A where:
For i = 1, · · · , r, Ai is isomorphic to Matni(∆i) for some ni and some division aaq-algebra
∆i.
For j = 1, · · · , s, Aj is isomorphic to Matnj ,mj(Dj) for some division algebra Dj and
natural numbers nj and mj.
A is a trivial unital aaq-algebras with semisimple even part. Moreover, r, s, the ni’s and
the pairs nj ,mj are uniquely determined by A; and so are (up to isomorphism) A, the
division aaq-algebras ∆i’s and the division algebras Dj’s.
In general if A = ⊕g∈GAg is a quasiassociative algebra over the field K , with simple
artinian null part B = A0, then it is isomorphic to V ⊗D∆⊗DV∗, where ∆ is a quasiassociative
division algebra, V is a simple A0−module and D = EndA0(V ). This is equivalent to the
following theorem [1],
Theorem 5.3. Any quasiassociative algebra A with simple artinian null part is isomorphic
to an algebra of matrices Matn(∆), for some integer n and quasiassociative division algebra
∆. The integer n is uniquely determined by A and so is, up to isomorphism, the division
algebra ∆.
13
6 Wedderburn quasialgebras
A Wedderburn quasialgebra is a unital quasialgebra that satisfies the descending chain con-
dition on graded left ideals and with no nonzero nilpotent graded ideals [3]. In this paper it
was proved an analogous of the Wedderburn-Artin Theorem for quasialgebras.
The first result obtained in [3] it was that if A is a Wedderburn quasialgebra, then the
null part A0 is a Wedderburn algebra.
Consider natural numbers n1, . . . , nr, define the matrix algebra Mn1+···+nr(k) having a
basis consisting of the elements
Ekl
ij= En1+···+ni−1+k,n1+···+nj−1+l (19)
where, as usual, Eij denotes the matrix with 1 in the (i, j) entry and 0’s elsewhere. The
multiplication is,
Ekl
ijEpq
mn= δjmδlpE
kq
in. (20)
Let R be a quasialgebra with cocycle φ. Consider in G a grading function |1|, . . . , |r| ∈ G
and take natural numbers n1, · · · , nr. Define the set of matrices,
Mφ
n1,··· ,nr(R) =< xkl
ij>=< Ekl
ijx >, : x ∈ R, 1 ≤ i, j ≤ r, 1 ≤ k ≤ ni, 1 ≤ l ≤ nj (21)
that is a G-graded algebra for the gradation,
|xkl
ij| = |i||x||j|−1 (22)
for homogeneous x, and multiplication given by
xkl
ijypq
mn= δjmδlp
φ(|i|, |x||j|−1, |j||y′||n|−1)φ(|x||j|−1, |j|, |y||n|−1)
φ(|x|, |j|−1, |j|)φ(|x|, |y|, |n|−1)(xy)kq
in. (23)
In fact, Mφ
n1,...,nr(R) is a quasialgebra with cocycle φ.
The main result in [3] is,
Theorem 6.1. Let A be a Wedderburn quasialgebra. Then A is isomorphic to a finite direct
sum of quasialgebras of the form Mφ
n1,...,nr(D), where D is a division quasi-algebra with cocycle
φ.
14
References
[1] H. Albuquerque and A. P. Santana. A note on quasiassociative algebras. In The J.A.
Pereira da Silva Birthday Schrift; Textos de Matematica, ser. B, 32:5–16, 2002.
[2] H. Albuquerque and A. P. Santana. Quasiassociative algebras with simple artinian null
part. Algebra Logic, 43(5):307–315, 2004.
[3] H. Albuquerque, A. Elduque, and J. M. Perez-Izquierdo. Wedderburn quasialgebras.
subbmited to publication.
[4] H. Albuquerque, A. Elduque, and J. M. Perez-Izquierdo. Alternative quasialgebras. Bull.
Austral. Math. Soc., 63(2):257–268, 2001.
[5] H. Albuquerque, A. Elduque, and J. M. Perez-Izquierdo. z2–quasialgebras. Comm.
Algebra, 30(5):2161–2174, 2002.
[6] H. Albuquerque and S. Majid. Quasialgebra structure of the octonions. J. Algebra,
220:188–224, 1999.
[7] H. Albuquerque and S. Majid. zn-quasialgebras. Textos de Mat. de Coimbra Ser. B,
19:57–64, 1999.
[8] H. Albuquerque and S. Majid. New approach to the octonions and cayley algebras. In
Marcel Dekker, editor, Lec. Notes Pure and Applied Maths, volume 211, pages 1–7. 2000.
[9] H. Albuquerque and S. Majid. Clifford algebras obtained by twisting of group algebras.
J. Pure Applied Algebra, 171:133–148, 2002.
[10] M. Cohen and S. Montgomery. Group graded rings, smash products and group actions.
Trans. Americ. Mat. Soc., 281:237–282, 1984.
[11] J. A. Cuenca-Mira, A. Garcıa-Martın, and C. Martın-Gonzalez. Prime associative su-
peralgebras with nonzero socle. Algebras Groups Geom., 11:359–369, 1994.
[12] G. Dixon. Division Algebras: Octonions, Quaternions, Complex Numbers and the Alge-
braic Design of Physics. Kluwer, 1994.
[13] C. Draper. Superalgebras asociativas. Tesina de Licenciatura, Universidad de Malaga,
1997.
[14] C. Draper and A. Elduque. Division superalgebras. In A. Castellon Serrano, J. A. Cuenca
Mira, A. Fernandez Lopez, and C. Martın Gonzalez, editors, Proceedings of the Interna-
tional Conference on Jordan Structures (Malaga 1977). Malaga, June 1999.
15
[15] V. G. Drinfeld. QuasiHopf algebras. Leningrad Math. J., 1:1419–1457, 1990.
[16] K. A. Zhevlakov et al. Rings that are nearly associative. Academic Press, New York and
London, 1982.
[17] C. Nastasescu. Graded ring theory. North-Holland, Amsterdam, 1982.
[18] R. Pierce. Associative algebras, volume 88 of Graduate texts in Mathematics. Springer,
1982.
[19] J. Racine. Primitive superalgebras with superinvolution. J. Algebra, 206:588–614, 1998.
[20] R. Schaffer. An Introduction to Nonassociative Algebras. Academic Press, New York and
London, 1966.
[21] M. E. Sweedler. Hopf Algebras. Benjamin, 1969.
[22] C. T. C. Wall. Graded brauer groups. J. Reine Angew. Math., 213:187–199, 1964.
16
Freudenthal’s magic supersquare in characteristic 3
Isabel Cunha∗ Alberto Elduque†
Abstract
A family of simple Lie superalgebras over fields of characteristic 3, with no counter-
part in Kac’s classification in characteristic 0 have been recently obtained related to an
extension of the classical Freudenthal’s Magic Square. This article gives a survey of these
new simple Lie superalgebras and the way they are obtained.
1 Introduction
Over the years, many different constructions have been given of the excepcional simple Lie
algebras in Killing-Cartan’s classification, involving some nonassociative algebras or triple
systems. Thus the Lie algebra G2 appears as the derivation algebra of the octonions (Cartan
1914), while F4 appears as the derivation algebra of the Jordan algebra of 3 × 3 hermitian
matrices over the octonions and E6 as an ideal of the Lie multiplication algebra of this Jordan
algebra (Chevalley-Schafer 1950).
In 1966 Tits [Tit66] gave a unified construction of the exceptional simple Lie algebras,
valid over arbitrary fields of characteristic 6= 2, 3 which uses a couple of ingredients: a unital
composition algebra (or Hurwitz algebra) C, and a central simple Jordan algebra J of degree
3:
T (C, J) = der C ⊕ (C0 ⊗ J0) ⊕ der J,
where C0 and J0 denote the sets of trace zero elements in C and J . By defining a suitable
Lie bracket on T (C, J), Tits obtained Freudenthal’s Magic Square ([Sch95, Fre64]):
∗Departamento de Matematica, Universidade da Beira Interior, 6200 Covilha, Portugal. Email:
[email protected]. Supported by Supported by Centro de Matematica da Universidade de Coimbra – FCT.†Departamento de Matematicas, Universidad de Zaragoza, 50009 Zaragoza, Spain. Email:
[email protected]. Supported by the Spanish Ministerio de Educacion y Ciencia and FEDER (MTM 2004-
081159-C04-02) and by the Diputacion General de Aragon (Grupo de Investigacion de Algebra)
17
T (C, J) H3(k) H3(k × k) H3(Mat2(k)) H3(C(k))
k A1 A2 C3 F4
k × k A2 A2 ⊕ A2 A5 E6
Mat2(k) C3 A5 D6 E7
C(k) F4 E6 E7 E8
At least in the split cases, this is a construction which depends on two unital composition
algebras, since the Jordan algebra involved consists of the 3 × 3 hermitian matrices over
a unital composition algebra. Even though the construction is not symmetric on the two
Hurwitz algebras involved, the result (the Magic Square) is symmetric.
Over the years, several symmetric constructions of Freudenthal’s Magic Square based on
two Hurwitz algebras have been proposed: Vinberg [Vin05], Allison and Faulkner [AF93] and
more recently, Barton and Sudbery [BS, BS03], and Landsberg and Manivel [LM02, LM04]
provided a different construction based on two Hurwitz algebras C, C ′, their Lie algebras of
triality tri(C), tri(C ′), and three copies of their tensor product: ιi(C ⊗ C ′), i = 0, 1, 2. The
Jordan algebra J = H3(C′), its subspace of trace zero elements and its derivation algebra can
be split naturally as:
J = H3(C′) ∼= k3 ⊕
(⊕2
i=0ιi(C′)),
J0∼= k2 ⊕
(⊕2
i=0ιi(C′)),
der J ∼= tri(C ′) ⊕(⊕2
i=0ιi(C′)),
and the above mentioned symmetric constructions are obtained by rearranging Tits construc-
tion as follows:
T (C, J) = der C ⊕ (C0 ⊗ J0) ⊕ der J
∼= der C ⊕ (C0 ⊗ k2) ⊕(⊕2
i=0C0 ⊗ ιi(C′))⊕(tri(C ′) ⊕ (⊕2
i=0ιi(C′)))
∼=(tri(C) ⊕ tri(C ′)
)⊕(⊕2
i=0ιi(C ⊗ C ′)
).
This construction, besides its symmetry, has the advantage of being valid too in charac-
teristic 3. Simpler formulas are obtained if symmetric composition algebras are used, instead
of the more classical Hurwitz algebras. This led the second author to reinterpret the above
construction in terms of two symmetric composition algebras [Eld04].
An algebra endowed with a nondegenerate quadratic form (S, ∗, q) is said to be a symmetric
composition algebra if it satisfies
q(x ∗ y) = q(x)q(y),
q(x ∗ y, z) = q(x, y ∗ z).
18
for any x, y, z ∈ S, where q(x, y) = q(x + y) − q(x) − q(y) is the polar of q.
Any Hurwitz algebra C with norm q, standard involution x 7→ x = q(x, 1)1 − x, but with
new multiplication
x ∗ y = xy,
is a symmetric composition algebra, called the associated para-Hurwitz algebra.
The classification of symmetric composition algebras was given by Elduque, Okubo, Os-
born, Myung and Perez-Izquierdo (see [EM93, EP96, KMRT98]). In dimension 1,2 or 4, any
symmetric composition algebra is a para-Hurwitz algebra, with a few exceptions in dimension
2 which are, nevertheless, forms of para-Hurwitz algebras; while in dimension 8, apart from
the para-Hurwitz algebras, there is a new family of symmetric composition algebras termed
Okubo algebras.
If (S, ∗, q) is a symmetric composition algebra, the subalgebra of so(S, q)3 defined by
tri(S) = (d0, d1, d2) ∈ so(S, q)3 : d0(x ∗ y) = d1(x) ∗ y + x ∗ d2(y)∀x, y ∈ S
is the triality Lie algebra of S, which satisfies:
tri(S) =
0 if dim S = 1,
2-dim’l abelian if dimS = 2,
so(S0, q)3 if dim S = 4,
so(S, q) if dim S = 8.
The construction given by Elduque in [Eld04] starts with two symmetric composition
algebras S, S′ and considers the Z2 × Z2-graded algebra
g(S, S′) =(tri(S) ⊕ tri(S′)
)⊕(⊕2
i=0ιi(S ⊗ S′)
),
where ιi(S ⊗ S′) is just a copy of S ⊗ S′, with anticommutative multiplication given by: tri(S) ⊕ tri(S′) is a Lie subalgebra of g(S, S′), [(d0, d1, d2), ιi(x ⊗ x′)] = ιi(di(x) ⊗ x′
), [(d′
0, d′
1, d′
2), ιi(x ⊗ x′)] = ιi
(x ⊗ d′
i(x′)
), [ιi(x ⊗ x′), ιi+1(y ⊗ y′)] = ιi+2
((x ∗ y) ⊗ (x′ ∗ y′)
)(indices modulo 3), [ιi(x ⊗ x′), ιi(y ⊗ y′)] = q′(x′, y′)θi(tx,y) + q(x, y)θ′i(t′
x′,y′),
19
for any x, y ∈ S, x′, y′ ∈ S′, (d0, d1, d2) ∈ tri(S) and (d′0, d′
1, d′
2) ∈ tri(S′). The triple tx,y =
(q(x, .)y−q(y, .)x, 1
2q(x, y)1−rxly,
1
2q(x, y)1−lxry
)is in tri(S) and θ : (d0, d1, d2) 7→ (d2, d0, d1)
is the triality automorphism in tri(S); and similarly for t′ and θ′ in tri(S′).
With this multiplication, g(S, S′) is a Lie algebra and, if the characteristic of the ground
field is 6= 2, 3, Freudenthal’s Magic Square is recovered.
dim S′
g(S, S′) 1 2 4 8
1 A1 A2 C3 F4
2 A2 A2 ⊕ A2 A5 E6
dimS4 C3 A5 D6 E7
8 F4 E6 E7 E8
In characteristic 3, some attention has to be paid to the second row (or column), where
the Lie algebras obtained are not simple but contain a simple codimension 1 ideal.
dim S′
g(S, S′) 1 2 4 8
1 A1 A2 C3 F4
2 A2 A2 ⊕ A2 A5 E6
dimS4 C3 A5 D6 E7
8 F4 E6 E7 E8 A2 denotes a form of pgl3, so [A2, A2] is a form of psl3. A5 denotes a form of pgl6, so [A5, A5] is a form of psl6. E6 is not simple, but [E6, E6] is a codimension 1 simple ideal.
The characteristic 3 presents also another exceptional feature. Only over fields of this
characteristic there are nontrivial composition superalgebras, which appear in dimensions 3
and 6. This fact allows to extend Freudenthal’s Magic Square with the addition of two further
rows and columns, filled with (mostly simple) Lie superalgebras.
A precise description of those superalgebras can be done as contragredient Lie superalge-
bras (see [CEa] and [BGL]).
Most of the Lie superalgebras in the extended Freudenthal’s Magic Square in character-
istic 3 are related to some known simple Lie superalgebras, specific to this characteristic,
constructed in terms of orthogonal and symplectic triple systems, which are defined in terms
of central simple degree three Jordan algebras.
20
2 The Supersquare and Jordan algebras
It turns out that the Lie superalgebras g(Sr, S1,2) and g(Sr, S4,2), for r = 1, 4 and 8, and their
derived subalgebras for r = 2, are precisely the simple Lie superalgebras defined in [Eld06b]
in terms of orthogonal and symplectic triple systems and strongly related to simple Jordan
algebras of degree 3.
Let S be a para-Hurwitz algebra. Then,
g(S1,2, S) =(tri(S1,2) ⊕ tri(S)
)⊕(⊕2
i=0ιi(S1,2 ⊗ S)
)
=((sp(V ) ⊕ V ) ⊕ tri(S)
)⊕(⊕2
i=0ιi(1 ⊗ S)
)⊕(⊕2
i=0ιi(V ⊗ S)
).
Consider the Jordan algebra of 3 × 3 hermitian matrices over the associated Hurwitz
algebra:
J = H3(S) =
α0 a2 a1
a2 α1 a0
a1 a0 α2
: α0, α1, α2 ∈ k, a0, a1, a2 ∈ S
∼= k3 ⊕(⊕2
i=0ιi(S)).
In [CEb, Theorem 4.9.] it is proved that:
g(S1,2, S)0∼= sp(V ) ⊕ der J (as Lie algebras),
g(S1,2, S)1∼= V ⊗ J (as modules for the even part).
Therefore, some results concerning Z2-graded Lie superalgebras and orthogonal triple
systems ([Eld06b]) allow us to conclude that J = J0/k1 is an orthogonal triple system with
product given by
[xyz] =(x (y z) − y (x z)
)ˆ
(x = x + k1).
Orthogonal triple systems were first considered by Okubo [Oku93].
Theorem 1 (Elduque-Cunha [CEb]). The Lie superalgebra g(S1,2, S) is the Lie superalgebra
associated to the orthogonal triple system J = J0/k1, for J = H3(S).
In order to analyze the Lie superalgebras g(S4,2, S), let V be, as before, a two dimensional
vector space endowed with a nonzero alternating bilinear form. Assume that the ground field
21
is algebraically closed. Then (see [CEb]) it can be checked that g(S8, S) can be identified to
a Z2-graded Lie algebra
g(S8, S) ≃(sp(V ) ⊕ g(S4, S)
)⊕(V ⊗ T (S)
),
for a suitable g(S4, S)-module T (S) of dimension 8 + 6dimS.
In a similar vein, one gets that g(S4,2, S) can be identified with
g(S4,2, S) ≃ g(S4, S) ⊕ T (S).
Hence, according to [CEb, Theorem 5.3, Theorem 5.6]:
Corollary 1 (Elduque-Cunha). Let S be a para-Hurwitz algebra, then:
1. T (S) above is a symplectic triple system.
2. g(S4,2, S) ∼= inder T ⊕ T is the Lie superalgebra attached to this triple system.
Remark 1. Symplectic triple systems are closely related to the so called Freudenthal triple
systems (see [YA75]). The classification of the simple finite dimensional symplectic triple
systems in characteristic 3 appears in [Eld06b, Theorem 2.32] and it follows from this clas-
sification that the symplectic system triple T (S) above is isomorphic to a well-known triple
system defined on the set of 2 × 2-matrices
(
k J
J k
)
, with J = H3(S).
The next table summarizes the previous arguments:
g S1 S2 S4 S8
S1,2 Lie superalgebras attached to orthogonal
triple systems J = J0/k1
S4,2 Lie superalgebras attached to symplectic
triple systems
(
k J
J k
)
(J a degree 3 central simple Jordan algebra)
22
3 Final remarks
In the previous section, the Lie superalgebras g(Sr, S1,2) and g(Sr, S4,2) (r = 1, 2, 4, 8) in the
extended Freudenthal’s Magic Square have been shown to be related to Lie superalgebras
which had been constructed in terms of orthogonal and symplectic triple systems in [Eld06b].
Let us have a look here at the remaining Lie superalgebras in the supersquare:
The result in [CEa, Corollary 5.20] shows that the even part of g(S1,2, S1,2) is isomorphic
to the orthogonal Lie algebra so7(k), while its odd part is the direct sum of two copies of
the spin module for so7(k), and therefore, over any algebraically closed field of characteristic
3, g(S1,2, S1,2) is isomorphic to the simple Lie superalgebra in [Eld06b, Theorem 4.23(ii)],
attached to a simple null orthogonal triple system.
For g(S4,2, S4,2), as shown in [CEa, Proposition 5.10 and Corollary 5.11], its even part is
isomorphic to the orthogonal Lie algebra so13(k), while its odd part is the spin module for
the even part, and hence that g(S4,2, S4,2) is the simple Lie superalgebra in [Elda, Theorem
3.1(ii)] for l = 6.
Only the simple Lie superalgebra g(S1,2, S4,2) has not previously appeared in the literature.
Its even part is isomorphic to the symplectic Lie algebra sp8(k), while its odd part is the
irreducible module of dimension 40 which appears as a subquotient of the third exterior
power of the natural module for sp8(k) (see [CEa, §5.5]).Also g(S1,2, S1,2) and g(S1,2, S4,2) are related to some orthosymplectic triple systems [CEb],
which are triple systems of a mixed nature.
In conclusion, notice that
the main feature of Freudenthal’s Magic Supersquare is that among all the Lie superal-
gebras involved, only g(S1,2, S1) ∼= psl2,2 has a counterpart in Kac’s classification in charac-
teristic 0. The other Lie superalgebras in Freudenthal’s Magic Supersquare, or their derived
algebras, are new simple Lie superalgebras, specific of characteristic 3.
References
[AF93] B. N. Allison and J. R. Faulkner, Nonassociative coefficient algebras for Steinberg
unitary Lie algebras, J. Algebra 161 (1993), no. 1, 1–19.
[BS] C. H. Barton and A. Sudbery, Magic Squares of Lie Algebras,
arXiv:math.RA/0001083.
[BS03] C. H. Barton and A. Sudbery, Magic squares and matrix models of Lie algebras,
Adv. Math. 180 (2003), no. 2, 596–647.
23
[BL06] Sofiane Bouarroudj and Dimitri Leites, Simple Lie superalgebras and nonintegrable
distributions in characteristic p, Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat.
Inst. Steklov. (POMI) 331 (2006), no. Teor. Predst. Din. Sist. Komb. i Algoritm.
Metody. 14, 15–29, 221.
[BGL] Sofiane Bouarroudj, Pavel Grozman and Dimitri Leites, Cartan matrices and
presentations of Cunha and Elduque superalgebras, arXiv:math.RT/0611391.
[Cun07] Isabel Cunha, Extensao do Quadrado Magico de Freudenthal em caracterıstica 3,
Doctoral Thesis, Universidade da Beira Interior.
[CEa] Isabel Cunha and Alberto Elduque, An extended Freudenthal Magic Square in
characteristic 3, arXiv:math.RA/0605379.
[CEb] , The extended Freudenthal Magic Square and Jordan Algebras,
arXiv:math.RA/0608191.
[Eld04] Alberto Elduque, The magic square and symmetric compositions, Rev. Mat.
Iberoamericana 20 (2004), no. 2, 475–491.
[Eld06a] , A new look at Freudenthal’s magic square, Non-associative algebra and its
applications, L. Sabinin, L.V. Sbitneva, and I. P. Shestakov, eds., Lect. Notes Pure
Appl. Math., vol. 246, Chapman & Hall/CRC, Boca Raton, FL, 2006, pp. 149–
165.
[Eld06b] , New simple Lie superalgebras in characteristic 3, J. Algebra 296 (2006),
no. 1, 196–233.
[Elda] , Some new simple modular Lie superalgebras, Pacific J. Math. (to appear),
arXiv:math.RA/0512654.
[Eldb] , The Magic Square and Symmetric Compositions II, Rev. Mat. Iberoamer-
icana (to appear), arXiv:math.RT/0507282.
[EM93] Alberto Elduque and Hyo Chul Myung, Flexible composition algebras and Okubo
algebras, Comm. Algebra 21 (1993), no. 7, 2481–2505.
[EO02] Alberto Elduque and Susumu Okubo, Composition superalgebras, Comm. Algebra
30 (2002), no. 11, 5447–5471.
[EP96] Alberto Elduque and Jose Marıa Perez, Composition algebras with associative
bilinear form, Comm. Algebra 24 (1996), no. 3, 1091–1116.
[Fre64] H. Freudenthal, Lie groups in the foundations of geometry, Adv. Math. 1 (1964),
145–190.
24
[Kac77] V. G. Kac, Lie superalgebras, Advances in Math. 26 (1977), no. 1, 8–96.
[KMRT98] Max-Albert Knus, Alexander Merkurjev, Markus Rost, and Jean-Pierre Tignol,
The book of involutions, American Mathematical Society Colloquium Publica-
tions, vol. 44, American Mathematical Society, Providence, RI, 1998.
[LM02] J. M. Landsberg and L. Manivel, Triality, exceptional Lie algebras and Deligne
dimension formulas, Adv. Math. 171 (2002), no. 1, 59–85.
[LM04] Joseph M. Landsberg and Laurent Manivel, Representation theory and projective
geometry, Algebraic transformation groups and algebraic varieties, Encyclopaedia
Math. Sci., vol. 132, Springer, Berlin, 2004, pp. 71–122.
[Oku93] Susumu Okubo, Triple products and Yang-Baxter equation. I. Octonionic and
quaternionic triple systems, J. Math. Phys. 34 (1993), no. 7, 3273–3291.
[Sch95] Richard D. Schafer, An introduction to nonassociative algebras, Dover Publica-
tions Inc., New York, 1995.
[Tit66] J. Tits, Algebres alternatives, algebres de Jordan et algebres de Lie exceptionnelles.
I. Construction, Nederl. Akad. Wetensch. Proc. Ser. A 69 = Indag. Math. 28
(1966), 223–237.
[Vin05] E.B. Vinberg, Construction of the exceptional simple Lie algebras., Vinberg,
Ernest (ed.), Lie groups and invariant theory. Providence, RI: American Math-
ematical Society (AMS). Translations. Series 2. American Mathematical Society
213. Advances in the Mathematical Sciences 56 (2005), 241–242.
[YA75] Kiyosi Yamaguti and Hiroshi Asano, On the Freudenthal’s construction of excep-
tional Lie algebras, Proc. Japan Acad. 51 (1975), no. 4, 253–258.
25
26
On graded Lie algebras and intersection cohomology
G. Lusztig∗
This is the text of a lecture given at the University of Coimbra in June 2006; a more
complete version will appear elsewhere.
1
Let G be a reductive connected group over C. Let LG be the Lie algebra of G. Let ι : C∗ −→
G be a homomorphism of algebraic groups. Let
Gι = g ∈ G; gι(t) = ι(t)g for all t ∈ C∗.
We have LG = ⊕n∈ZLnG where
LnG = x ∈ LG; Ad(ι(t))x = tnx ∀t ∈ k∗.
Now Gι acts on LkG by the adjoint action. If k 6= 0 this action has finitely many orbits. Fix
∆ = n,−n
where n ∈ Z − 0. For n ∈ ∆ let ILnG be the set of all isomorphism classes of irreducible
Gι-equivariant local systems on various Gι-orbits in LnG. For L,L′ in ILnG (on the orbits
O,O′) and i ∈ Z let mi;L,L′ be the multiplicity of L′ in the local system obtained by restricting
to O′ the i-th cohomology sheaf of the intersection cohomology complex IC(O,L).
Let mL,L′ =∑
imi,L,L′vdimO−dimO′−〉. These form the entries of the multiplicity matrix.
The problem is to compute it.
Example 1. Let V be a vector space of dimension 4 with a nonsingular symplectic form.
Let G = Sp(V ). Let V1, V−1 be Lagangian subspaces of V such that V = V1 ⊕ V−1. Let
∆ = 2,−2. Define ι : C∗ −→ G by ι(t)x = tx if x ∈ V1, ι(t)x = t−1x if x ∈ V−1.
We have L2G = S2V1, L−2V = S2(V ∗1) and IL2G consists of 4 local systems
L′, C on the orbit of dimension 0;
L∈, C on the orbit of dimension 2;
L∋, C on the orbit of dimension 3;
L, non-trivial on the orbit of dimension 3.
∗Department of Mathematics, M.I.T., Cambridge, MA 02139
27
Example 2. Let V be a vector space of dimension d. Let G = GL(V ). Assume that V is
Z-graded: V = ⊕k∈ZVk. Define ι : C∗ −→ G by ι(t)x = tkx for x ∈ Vk. Take ∆ = 1,−1.
Let L1G = T ∈ End(V );TVk ⊂ Vk+1∀k. The Gι-orbits on L1G are in bijection with the
isomorphism classes of representations of prescribed dimension of a quiver of type A. In this
case our problem is equivalent to the problem of describing the transition matrix from a PBW
basis to the canonical basis of a quantized enveloping algebra U+ of type A.
This suggests that to solve our problem we must immitate and generalize the construction
of canonical bases of U+v
.
2
Let B be the canonical basis of U+v
the plus part of a quantized enveloping algebra of finite
simply laced type.
B was introduced in the author’s paper in J.Amer.Math.Soc. (1990) by two methods: topological: study of perverse sheaves on the moduli space of representations of a quiver. algebraic: define a Z[v]-lattice L in U+v
and a Z- basis B0 of L/vL (basis at v = 0) in
terms of PBW-bases then lift B0 to B.
Another proof of the existence of B was later given by Kashiwara (Duke Math.J., 1991).
3
Let n ∈ ∆. Let L ∈ ILnG on an orbit O. We say that L is ordinary if there exists a
G-equivariant local system F on C (the unique nilpotent G-orbit in LG containig O) such
that
L appears in F|O and (C,F) appears in the Springer correspondence for G.
It is known that for L,L′ ∈ IL\G we have
L ordinary, mL,L′ 6= 0 implies L′ ordinary;
L′ ordinary, mL,L′ 6= 0 implies L ordinary.
Let Iord
LnG= L ∈ ILnG;L ordinary.
We say that (G, ι) is rigid if there exists a homomorphism γ : SL2(C) −→ G which maps
the diagonal matrix with diagonal entries tn, t−n to ι(t2)×(centre of G) for any t ∈ C∗.
Definition. Let K(LnG) be the Q(v)-vector space with two bases (L), (L) indexed by
Iord
LnGwhere L =
∑
L′ mL,L′L′.
4
Let U+v
be as in §2. Let (ei)i∈I be the standard generators of U+v
. Then U+v
is the quotient
of the free associative Q(v)-algebra with generators ei by the radical of an explicit symmetric
28
bilinear form on it. We have
U+
v= ⊕νU
+
v,ν
where U+
v,νis spanned by the monomials
ec1
i1ec2
i2. . . ecr
ir
with∑
s;is=ics = ν(i) for any i ∈ I. Thus U+
vis the quotient of the Q(v)-vector space
spanned by ec1
i1ec2
i2. . . ecr
iras above by the radical of an explicit symmetric bilinear form on it.
5 Combinatorial parametrization of set of Gι-orbits in LnG
Let P be the set of parabolic subgroups of G. Let
Pι = P ∈ P; ι(C) ⊂ P.
If P ∈ P we set P = P/UP ; we have ι : C∗ −→ P naturally. Let n ∈ ∆. Let P ∈ Pι be such
that (P , ι) is rigid. Choose a Levi subgroup M of P such that ι(C∗) ⊂ M . Let s ∈ [LM,LM ]
be such that [s, x] = kx for any k ∈ Z, x ∈ LkM . (Note that s is unique.) Let
LrG = x ∈ LG; [s, x] = rx,
Lr
tG = LrG ∩ LtG.
Then
LG = ⊕r∈(n/2)Z;s∈ZLr
tG.
We say that P is n-good if
LUP = ⊕r∈(n/2)Z;s∈Z;2t/n<2r/nLr
tG.
Then automatically
LM = ⊕r∈(n/2)Z;s∈Z;2t/n=2r/nLr
tG.
Let
Pn = P ∈ Pι; (P , ι) is rigid, P is n-good,
Pn
= Gι\Pn.
We claim that
Pn
∼−→ Gι − orbits in LnG.
Let P,Lr
tG be as above. Then L0
0G = LG′ where G′ acts by Ad on Ln
nG. Let O′ be the
unique G′-orbit on Ln
nG. Now G′ ⊂ Gι hence there is a unique Gι-orbit O in LnG containing
O′. The bijection above is given by P 7→ O. Other parametrizations of Gι − orbits in LnG
were given by Vinberg (1979), Kawanaka (1987).
29
6
Let B be the variety of Borel subgroups of G. Let Bι = B ∩ Pι. Let IG = B;B ∈ Bι. Let
JG
= Gι\JG . Let KG be the Q(v)-vector space with basis (IS)S∈JG. For Q ∈ Pι we define
a linear map (induction)
indG
Q−→ K
Q−→ KG
by
IQι−orbit of B′ 7→ IGι−orbit of B
where B is the inverse image of B′ under Q −→ Q. Let¯: Q(v) −→ Q(v) be the Q-linear map
such that vm = v−m for m ∈ Z. Define a ”bar operation” β : KG −→ KG by β(fIS) = f IS .
Inspired by the results in §4 we define a bilinear pairing
(:) : KG × KG −→ Q(v)
by
(IS : IS′) =∑
Ω
(−v)τ(Ω)
where Ω runs over the Gι-orbits on Bι ×Bι such that pr1Ω = S, pr2Ω = S ′ and if (B,B′) ∈ Ω
we set
τ(Ω) = − dimL0U
′ + L0U
L0U ′ ∩ L0U+ dim
LnU ′ + LnU
LnU ′ ∩ LnU;
here U,U ′ are the unipotent radicals of B,B′. Let R be the radical of (:). Let KG = KG/R.
Then (:) induces a nondegenerate bilinear pairing (:) on KG. Note that β induces a map
KG −→ KG denoted again by β. Moreover indG
Qinduces a map K
Q−→ KG denoted again
by indG
Q.
For n ∈ ∆ and η ∈ Pn
we define subsets Zη
n of KG by induction on dimG.
Assume that η ∈ Pn, η 6= G. We set Z
η
n = indG
P(Z P
n) where P ∈ η.
We set Z ′n
= ∪η∈Pn;η 6=GZη
n. Then:
the last union is disjoint and indG
P: ZP −→ Z
η
n is a bijection.
For η ∈ Pn
and P ∈ η we set dη = dim L0G− dim L0P + dim LnP . Define a partial order
≤ on Pn − G by
η′ < η if and ony if dη′ < dη.
η′ ≤ η if and only if η = η′ or η′ < η.
If ξ ∈ Z ′n
we have β(ξ) =∑
ξ′∈Z′n
aξ,ξ1ξ1 where aξ,ξ1
∈ Z[v, v−1] are unique and
aξ,ξ16= 0 implies η1 < η or ξ = ξ1 (here ξ ∈ Z
η
n, ξ1 ∈ Zη1
n )
aξ,ξ1= 1 if ξ1 = ξ.
Also,∑
ξ2∈Z′n
aξ,ξ2aξ2,ξ1
= δξ,ξ1for ξ, ξ1 ∈ Z ′
n. By a standard argument there is a unique
family of elements cξ,ξ1∈ Z[v] defined for ξ, ξ1 in Z ′
nsuch that
cξ,ξ1=
∑
ξ2∈Z′n
cξ,ξ2aξ2,ξ1
;
cξ,ξ16= 0 implies η1 < η or ξ = ξ1 (where ξ ∈ Z
η
n, ξ1 ∈ Zη1
n )
30
cξ,ξ16= 0, ξ 6= ξ1 implies cξ,ξ1
∈ vZ[v];
cξ,ξ1= 1 if ξ = ξ1.
For ξ ∈ Z ′n
we set Wξ
n =∑
ξ1∈Z′n
cξ,ξ1ξ1. Then β(W ξ
n) = Wξ
n.
Define Yn : KG −→ KG by (Yn(x) : Z ′n) = 0 and x = Yn(x) +
∑
ξ∈Z′n
Y γξξ where
γξ ∈ Q(v). This is well defined since the matrix ((ξ, ξ′))ξ,ξ′∈Z′n
is invertible. Let
J−n = ξ0 ∈ Z ′−n
;Yn(W ξ0
−n) 6= 0.
Let Zn = Z ′n
if (G, ι) is not rigid. If (G, ι) is rigid let
Zn = Z ′n∪ ξ; ξ = Yn(W ξ0
−n) for some ξ0 ∈ J−n.
We say that Zn a PBW-basis of KG. It depends on n. For ξ ∈ Zn we define Wξ
n as follows.
Wξ
n is as above if ξ ∈ Z ′n. W
ξ
n = Wξ
−nwhere ξ = Yn(W ξ0
−n), ξ0 ∈ J−n. We say that (W ξ
n) is
the canonical basis of KG. It does not depend on n.
Theorem. The transition matrix expressing the canonical basis in terms of the PBW basis
Zn coincides with the multiplicity matrix (mL,L′) in §1 (with L,L′ ordinary). There is an
appropriate generalization for not necessarily ordinary L,L′.
31
32
On the quantization of a class of non-degenerate
triangular Lie bialgebras over IK[[~]]
Carlos Moreno∗ Joana Teles†
Abstract
Let IK be a field of characteristic zero. Let IK[[~]] be the principal ideal domain of
formal series in ~ over IK. Let (a, [, ]a, εa = dcr1) be a finite dimensional non-degenerate
triangular Lie-bialgebra over IK. Let (a~, [, ]a~, εa~
= dc(~)r1(~)) be a deformation non-
degenerate triangular Lie-bialgebra over IK[[~]] corresponding to (a, [, ]a, εa = dcr1). The
aim of this talk is
a) To quantize (a~, [, ]a~, εa = dc(~)r1(~)) in the framework by Etingof-Kazhdan for
the quantization of Lie-bialgebras. Let Aa~ , J
−1r1(~)
be the Hopf Q.U.E. algebra over
IK[[~]] so obtained.
b) Let (a~, [, ]a~, εa~
= dc(~)r′1(~)) be another deformation non-degenerate triangular
Lie-bialgebra corresponding to (a, [, ]a, εa~= dcr
′
1). Let Aa~ , J
−1
r′
1(~)
be its quanti-
zation. Let β1(~), β′
1(~) ∈ a∗
~⊗IK[[~]] a
∗
~be the corresponding 2-forms associated
respectively to r1(~) and r′1(~). We prove that J−1r1(~) and J−1
r′
1(~) are equivalent in
the Hochschild cohomology of the universal enveloping algebra Ua~ if and only if
β1(~) and β′
1(~) are in the same class in the Chevalley cohomology of the Lie algebra
(a~, [, ]a~) over IK[[~]].
Keywords: Quantum Groups, Quasi-Hopf algebras, Lie bialgebras.
1 Some definitions
1.1 Lie bialgebras and Manin triples
1.1.1 Lie bialgebras over IK
Let IK be a field of characteristic 0. Let IK[[~]] = IK~ be the ring of formal power series in ~
with coefficients in IK. It is a principal ideal domain (PID) and its unique maximal ideal is
~IK~.
Let (a, [, ]a) be a finite dimensional Lie algebra over IK.
∗Departamento de Fısica Teorica, Universidad Complutense, E-28040 Madrid, Spain.†CMUC, Department of Mathematics, University of Coimbra, 3001-454 Coimbra, Portugal. E-mail:
33
A finite dimensional Lie bialgebra over IK is a set (a, [, ]a, IK, εa) where εa : a → a ⊗ a is
a 1-cocycle of a, with values in a ⊗ a, with respect to the adjoint action of a such that
εta : a∗ ⊗ a
∗ → a∗ is a Lie bracket on a
∗. It is called quasi-triangular if εa = dcr1, where dc is the Chevalley-Eilenberg coboundary,
r1 ∈ a ⊗ a is a solution to CYBE ([r1, r1] = 0) and (r1)12 + (r1)21 is ada-invariant. In case r1 is skew-symmetric, it is said to be a triangular Lie bialgebra. Moreover if
det(r1) 6= 0 it is called a non-degenerate triangular Lie bialgebra.
1.1.2 Deformation Lie bialgebras
Consider now the IK[[~]]-module obtained from extension of the scalars IK[[~]] ⊗IK a.
Let (a~ = IK[[~]] ⊗IK a = a[[~]], [, ]a~, IK~ = IK[[~]], εa~
), where εa~: a~ → a~ ⊗IK~
a~ is a
1-cocycle of a~ with values in tha adjoint representation, be a Lie bialgebra over IK[[~]].
It is a deformation Lie bialgebra of the Lie bialgebra over IK, a.
1.1.3 Manin triples
A Manin triple over IK~ is a set (g~ = (g~)+ ⊕ (g~)−, [, ]g~, <;>g~
) where (g~, [, ]g~) is a (finite dimensional) Lie algebra over IK~; ((g~)±, [, ](g~ )±
, IK~
)is a (finite dimensional) Lie algebra over IK~; <;>g~
is a non-degenerate, symmetric, bilinear, adg~-invariant form on g~; [, ]g~
|(g~)±= [, ](g~ )±
.
1.1.4 From Manin triples to Lie bialgebras
If (g~ = (g~)+ ⊕ (g~)−, [, ]g~, <;>g~
) is a Manin triple then There is a IK~-linear isomorphism χ−1 : (g~)− −→ (g~)∗+
because
<;>g~is non-degenerate. We may define in g~ = (g~)+ ⊕ (g~)∗+ a structure of a Manin triple. There is a structure of Lie bialgebra on (g~)+.
1.1.5 From Lie bialgebras to Manin triples
Let (a~, [, ]a~, IK~, εa~
) be a Lie bialgebra over IK~. Then (g~ = a~ ⊕ a∗~, [, ]a~⊕a∗
~, 〈; 〉a~⊕a∗
~),
where
[(x; ξ), (y; η)]g~= ([x, y]a~
+ ad∗ξy − ad∗
ηx; [ξ, η]a∗
~+ ad∗
xη − ad∗
yξ),
34
is a Manin triple.
The set (g~ = a~ ⊕ a∗~, [, ]a~⊕a∗
~, εg~
= dc(~)r), where r ∈ g~ ⊗IK~g~ (canonical), is a
quasi-triangular Lie bialgebra with [, ]g∗~
=(
[, ]a∗~;−[, ]a~
)
.
It is called the (quasi-triangular Lie bialgebra) classical double of the Lie bialgebra (a~, [, ]a~, IK~, εa~
).
1.1.6 Drinfeld Theorem
Drinfeld quasi-Hopf quasi-triangular QUE algebra corresponding to the deformation Lie al-
gebra (g~ = a~ ⊕ a∗~, [, ]g~
, r = a+ t ∈ g~ ⊗IK~g~ invariant) is
(Ug~, ·, 1,∆g~, ǫg~
,Φg~= eP (~t
12;~t
23), Sg~
, α = c−1, β = 1, Rg~= e
~
2t)
where c =∑
iXiSg~
(Yi)Zi, Φg~=∑
iXi⊗Yi⊗Zi and P is a formal Lie series with coefficients
in IK (or just in Q.)
2 Etingof-Kazhdan quantization
2.1 The classical double
2.1.1 Quantization of the double
Given the bialgebra (g~ = a~⊕a∗~, [, ]a~⊕a∗
~, εg~
= dc(~)r) double of the Lie bialgebra (a~, [, ]a~, IK~, εa~
),
we will construct a twist of Drinfeld quasi-Hopf QUE algebra
J ∈ Ug~⊗IK~Ug~
obtaining a Hopf QUE algebra. To do it, we need the following elements.
2.1.2 Drinfeld category
The category Mg~is defined as ObMg
~= topologically free g~-modules, that is X ∈ ObMg
~iff
1. X = V [[~]], where V is a vector space over IK
2. V [[~]] is a g~-module Of course, ObMg~
= topologically free Ug~-modules HomMg~(U [[~]], V [[~]]) is the set of g~-module morphisms, U [[~]]
f
−→ V [[~]]
1. f(a(~).x(~)) = a(~).f(x(~)), a(~) ∈ g~
2. f((a(~) b(~) − b(~) a(~))x(~)) = [a(~), b(~)]g~f(x(~))
35
and f extends in a unique way to a U(g~)-module morphism.
Theorem 2.1.1. Homg~(U [[~]], V [[~]]) is a torsion free IK~-module.
In the topology defined by its natural filtration
~pHomg~
(U [[~]], V [[~]])
its completion Homg~(U [[~]], V [[~]]) is separated, complete and a torsion free IK~-module.
It is then (see Kassel) a topologically free IK~-module, so
Homg~(U [[~]], V [[~]]) = Homa⊕a∗(U, V )[[~]].
The canonical element r ∈ g~⊗IK~g~ defines ΩVi(~)Vj(~) ∈ EndIK~
(V1[[~]]⊗V2[[~]]⊗V3[[~]]),
i, j = 1, 2, 3; i 6= j by
ΩVi(~)Vj (~)(v1(~) ⊗ v2(~) ⊗ v3(~)) = (· · ·i
fk ⊗ · · · ⊗
j
fk ⊗ · · ·+
+ · · ·
i
fk ⊗ · · · ⊗j
fk ⊗ · · · ).(v1(~) ⊗ v2(~) ⊗ v3(~)).
Lemma 2.1.2. ΩVi(~)Vj (~) ∈ Homg~(V1[[~]] ⊗ V2[[~]] ⊗ V3[[~]], V1[[~]] ⊗ V2[[~]] ⊗ V3[[~]]).
2.1.3 Tensor structure on Mg~
The element r is adg~-invariant. The Lie associator Φg~
∈ Ug~⊗Ug~⊗Ug~ is also invariant;
that is
Φg~· (∆g~
⊗ 1)∆g~(x(~)) = (1 ⊗ ∆g~
)∆g~(x(~)) · Φg~
, x(~) ∈ g~.
¿From this equality and the fact that Φg~= expP (~t12,~t23) we may define a g~-morphism
ΦV1(~),V2(~),V3(~) ∈ HomUg~
((V1(~)⊗V2(~))⊗V3(~) ,
V1(~)⊗(V2(~)⊗V3(~))).
Theorem 2.1.3. ΦV1(~),V2(~),V3(~) is an isomorphism in category Mg~and the set of these
isomorphisms defines a natural isomorphism between functors:
⊗(⊗× Id) −→ ⊗(Id×⊗).
2.1.4 Braided tensor structure on Mg~
For any couple
V1[[~]], V2[[~]] ∈ ObMg~
consider the isomorphism
βV1(~)V2(~) : V1(~)⊗V2(~) −→ V2(~)⊗V1(~)
u(~)⊗v(~) 7−→ σ
(
e~
2Ω12(~)u(~)⊗v(~)
)
,
36
where σ is the usual permutation.
Then βV1(~)V2(~) ∈ HomUg~
(V1(~)⊗V2(~), V2(~)⊗V1(~)
).
Theorem 2.1.4. The set of isomorphisms βV1(~)V2(~) of Mg~defines a natural isomorphism
between functors ⊗ −→ ⊗ σ. Then the category Mg~has a braided tensor structure.
2.1.5 The category A ObA = topologically free IK~-modules HomA(V1[[~]], V2[[~]]) = f : V1[[~]] −→ V2[[~]] , IK~-linear maps preserving filtrations
(⇐⇒ continuous)
Lemma 2.1.5. The category A is a strict monoidal symmetric category:
Φ = Id :(V [[~]]⊗U [[~]]
)⊗W [[~]] ≃ V [[~]]⊗
(U [[~]]⊗W [[~]]
),
and
σ : V [[~]]⊗U [[~]] −→ U [[~]]⊗V [[~]]
u(~) ⊗ v(~) 7−→ v(~) ⊗ u(~).
2.1.6 The functor F
Let F be the following map
F : Mg~−→ A
where F(V [[~]]) = HomUg~
(Ug~, V [[~]])
= Homa⊕a∗=g0(Ug0, V )[[~]] for f ∈ Homg~
(V [[~]], U [[~]]), then
F(f) ∈ HomA
(
Homg~(Ug~, V [[~]]),Homg~
(Ug~, U [[~]]))
is defined as (F(f))(g) = f g ∈ F(U [[~]]) ∈ ObA, and g ∈ F(V [[~]]).
Then F is a functor.
2.1.7 Tensor structure on F
We should now equip this functor F with a tensor structure. We will use the decomposition
g~ = a~ ⊕ a∗~ = (g~)+ ⊕ (g~)−
to produce such a structure.
37
2.1.8 g~-modules M(~)+ and M(~)− By Poincare-Birkhoff-Witt theorem we have IK~-module isomorphisms
– U(g~)+ ⊗IK~U(g~)− −→ Ug~ (product in Ug~)
– U(g~)− ⊗IK~U(g~)+ −→ Ug~ (product in Ug~) Let the IK~-modules of rang 1
W±(~) = a(~).e± : a(~) ∈ IK~, e± basis
We endow W± with a trivial U(g~)±-module structure.
– x(~)±(a(~)e±) = a(~)(x(~)±e±) = 0±, x(~)± ∈ U(g~)± Define the corresponding induced left Ug~-modules:
M(~)± = Ug~ ⊗U(g~)±W±(~) = · · · = U(g~)∓ · 1±
where 1± = (1 ⊗U(g~)±e±), 1 is the unit of Ug~.
As IK~-modules M(~)± is then the IK~-module U(g~)∓.
Lemma 2.1.6. M(~)± ∈ ObMg~.
Theorem 2.1.7. There exists a unique g~-module morphism
i± : M(~)± −→M(~)± ⊗IK~M(~)±
such that
i±(1±) = 1± ⊗IK~1±.
It is continuous for the (~)-adic topology and extends in a unique way to a g~-module morphism
i± : M(~)± −→ M(~)±⊗IK~M(~)±.
Theorem 2.1.8. Define the following g~-module morphism
φ : U(g~) −→M(~)+ ⊗IK~M(~)−
by φ(1) = 1+ ⊗ 1−.
Then φ is an isomorphism of g~-modules.
Proof. If φ(1) = 1+ ⊗ 1− and φ is a g~-module morphism, the construction of φ is unique.
φ preserves the standard filtration, then it defines a map gradφ on the associated graded
objects. This map gradφ is bijective and then (Bourbaki) φ is bijective.
38
2.1.9 Natural isomorphism of functors J
Definition 2.1.9. A tensor structure on the functor F : Mg~−→ A is a natural isomorphism
of functors
J : F(·) ⊗F(·) −→ F(· ⊗ ·)
We define this tensor structure following the same pattern that in Etingof-Kazhdan.
For V~ = V [[~]], W~ = W [[~]] ∈ ObMg~, define
JV~ ,W~: F(V~)⊗IK~
F(W~) −→ F(V~⊗W~)
v~ ⊗ w~ −→ JV~ ,W~(v~ ⊗ w~),
where v~ ∈ F(V~) = Homg~(Ug~, V~) and w~ ∈ F(W~), then
JV~ ,W~(v~ ⊗ w~) = (v~ ⊗ w~) (φ−1 ⊗ φ−1) Φ−1
M(~)+,M(~)−,M(~)+⊗M(~)−
(1 ⊗ ΦM(~)−,M(~)+,M(~)−) (1 ⊗ (σ e
~
2ΩM(~)+,M(~)− ) ⊗ 1)
(1 ⊗ Φ−1
M(~)+,M(~)−,M(~)−) ΦM(~)+,M(~)+,M(~)−⊗M(~)−
(i+ ⊗ i−) φ.
2.1.10 Definition of J
Theorem 2.1.10. The maps JV~ ,W~are isomorphisms and define a tensor structure on the
functor F .
From this tensor structure we get as in the IK case the following element in Ug~⊗Ug~
J = (φ−1 ⊗ φ−1)[(
Φ−1
1,2,34· Φ2,3,4 · e
~
2Ω23 · (σ23Φ
−1
2,3,4)·
· (σ23Φ1,2,34)) (1+⊗1−⊗1+⊗1−)] .
2.1.11 Quantization of the classical double
Theorem 2.1.11. The set (Ug~, ·, 1,∆′, ǫg~
, S′, R′), such that
∆′(u~) = J−1 · ∆g~(u~) · J
S′(u~) = Q−1(~) · Sg~(u~) ·Q(~),
R′ = σJ−1 · e~
2Ω · J,
where Q(~) is an element in Ug~ obtained from J and Sg~, is a quasi-triangular Hopf algebra,
and it is just the algebra obtained twisting via the element J−1 the quasi-triangular quasi-
Hopf QUE algebra obtained in Subsection 1.1.6.
In particular,
Φ · (∆g~⊗ 1)J · (J ⊗ 1) = (1 ⊗ ∆g~
)J · (1 ⊗ J),
39
J = 1 +~
2r +O(~2)
R = 1 + ~r +O(~2)
We call the above quasi-triangular Hopf quantized universal enveloping algebra a quanti-
zation of the quasi-triangular Lie bialgebra over IK~
(g~ = a~ ⊕ a∗~, [, ]g~
, εg~= dc(~)r).
double of the Lie bialgebra (a~, [, ]a~, εa~
) over the ring IK~.
2.2 Quantization of Lie bialgebras
2.2.1 Quasi-triangular Lie bialgebras
If the IK~-Lie bialgebra is quasi-triangular, that is, if εa~is an exact 1-cocycle
εa~= dc(~)r1(~), r1(~) ∈ a~ ⊗ a~
and [r1(~), r1(~)]a~= 0, we want to obtain a quantization of the IK~-Lie bialgebra
(a~, [, ]a~, εa~
= dc(~)r1(~)).
Following Etingof and Kazhdan and also an idea by Reshetikhin and Semenov-Tian-
Shansky, we will construct a Manin triple over IK~ and quantize it as we did before.
Lemma 2.2.1. There exist basis ai(~), i = 1, . . . , dim a and bj(~), j = 1, . . . , dim a of
the IK~-module a~ such that
r1(~) =
l∑
i=1
ai(~) ⊗ bi(~) ∈ a~ ⊗IK~a~. (2.1)
Proof. As IK~ is a principal ideal domain (PID) and a~ is a free IK~-module, r1(~) has a rang,
l, and a theorem about matrices with entries in a PID-module asserts that basis verifying
(2.1) exist.
Let us define maps, as in Reshetikhin-Semenov,
µr1(~), λr1(~) : a∗~ −→ a~
as
λr1(~)(f) =∑
ai(~).f(bi(~))
µr1(~)(f) =∑
f(ai(~)).bi(~), f ∈ a∗~
and write
(a~)+ = Im λr1(~) (a~)− = Im µr1(~).
40
We have
dimIK~(a~)+ = dimIK~
(a~)− = l = rang r1(~)
We may prove as in the IK case.
Lemma 2.2.2. The mapping
χr1(~) : (a~)∗+−→ (a~)−
g 7−→ χr1(~)(g) = (g ⊗ 1)r1(~)
is a IK~-module morphism (IK~ is a PID and a~ is free over IK~).
(a~)+ and (a~)− are Lie subalgebras of (a~, [, ]a~)
On the IK~-module g~ = (a~)+⊕ (a~)− we may define a skew-symmetric, bilinear mapping
such that [, ]g~|(a~)±
= [, ](a~)±.
Theorem 2.2.3. Let π be defined as
π : g~ = (a~)+ ⊕ (a~)− −→ a~
(x(~); y(~)) −→ x(~) + y(~).
Then,
(a) π([(x(~); y(~)), (z(~);u(~))](a~ )+⊕(a~)−
)=
= [π(x(~); y(~)), π(z(~);u(~))]a~;
(b) ((a~)+ ⊕ (a~)−, [, ](a~)+⊕(a~)−, IK~) is a Lie algebra over IK~.
¿From (a) and (b), π is a Lie algebra morphism.
Theorem 2.2.4. The set (g~ = (a~)+ ⊕ (a~)−, [, ]g~, 〈; 〉g~
) where
〈(x+(~);x−(~)); (y+(~); y−(~))〉g~= χ−1
r1(~)(x−(~)) .y+(~) + χ−1
r1(~)(y−(~)) .x+(~),
x+(~), y+(~) ∈ (a~)+, x−(~), y−(~) ∈ (a~)−, is a Manin triple.
In particular, the 2-form 〈; 〉g~is adg~
-invariant.
Because it is a Manin triple the set
((a~)+, [, ](a~ )+, ε)
where ε : (a~)+ −→ (a~)+ ⊗ (a~)+ is the transpose of the Lie bracket on (a~)∗+ defined as
[ξ(~), η(~)](a~ )∗+
= χ−1
r1(~)
([χr1(~)(ξ(~)), χr1(~)(η(~))](a~ )−
),
is a Lie bialgebra.
41
Definition 2.2.5. Let ei(~), i = 1, . . . , l a basis of (a~)+ and ei(~), i = 1, . . . , l its dual
basis on (a~)∗+. Let
r(~) = (ei(~); 0) ⊗ (0; ei(~))
the canonical element in the IK~-module (a~)+ ⊕ (a~)∗+. Define
r(~) = (1 ⊕ χr1(~)) ⊗ (1 ⊕ χr1(~)).r(~) ∈ (a~)+ ⊕ (a~)−
We prove
Theorem 2.2.6. The set (g~ = (a~)+⊕(a~)−, [, ]g~, εg~
= dc(~)r(~)) is a quasi-triangular Lie
bialgebra which is isomorphic to the quasi-triangular Lie bialgebra (g~ = (a~)+⊕(a~)∗+, [, ]g~
, εg~=
dc(~)r(~)), double of the Lie bialgebra ((a~)+, [, ](a~ )+, ε).
Theorem 2.2.7. Let the mapping π be defined by the commutativity of the diagram
- a~
6
(a~)+ ⊕ (a~)∗+
g~ = (a~)+ ⊕ (a~)−
(1 ⊕ χr1(~))
π
π
So, π = π (1 ⊕ χr1(~)).
Then, π is a Lie bialgebra homomorphism verifying
(π ⊗ π)r(~) = r1(~).
2.2.2 Quantization of the quasi-triangular Lie bialgebra
(a~, [, ]a~, εa~
= dc(~)r1(~), [r1(~), r1(~)]a~= 0, r1(~) ∈ a~ ⊗ a~)
We may project using the Lie bialgebra morphism
π : g~ = (a~)+ ⊕ (a~)∗+−→ a~
what we have done, about quantization of g~. We will get a quantization of a~. Precisely,
Theorem 2.2.8. Let (Ua~, ·, 1,∆a~, ǫa~
, Sa~) be the usual Hopf algebra. Let
(Ug~, ·, 1,∆g~, ǫg~
,Φg~, Sg~
, Rg~= e
~
2Ω)
be the quasi-triangular quasi-Hopf algebra in Subsection 1.1.6.
Let Φa~= (π ⊗ π ⊗ π)Φg~
and Ra~= (π ⊗ π)Rg~
, then
42
(π ⊗ π) ∆g~= ∆a~
π π Sg~= Sa~
π The set
(Ua~, ·, 1,∆a~, ǫa~
, Φa~, Sa~
, Ra~)
is a quasi-triangular quasi-Hopf QUE algebra. We call it a quantization (quasi-Hopf
one) of the Lie bialgebra (a~, r1(~) + σr1(~)).
Let (Ug~, ·, 1,∆′g~, ǫg~
, S′g~, R′
g~) where ∆′
g~(u~) = J−1
g~· ∆g~
(u~) · Jg~ S′g~
(u~) = Q−1(~) · Sg~(u~) ·Q(~) R′
g~= σJ−1
g~· e
~
2Ω · Jg~
be the quasi-triangular Hopf QUE algebra obtained before following E-K scheme of quanti-
zation of the classical double
(g~ = (a~)+ ⊕ (a~)∗+, [, ]g~, εg~
= dc(~)r).
Let us define Ja~= (π ⊗ π)Jg~
; ∆a~(a~) = J−1
a~· ∆a~
(a~) · Ja~ Ra~= (π ⊗ π)R′
g~; ǫa~
the usual counit in U(a~) Sa~(a~) = Q−1(~)·Sa~
(a~)·Q(~) where Q =∑Sa~
(ri(~)).si(~) and Ja~=∑ri(~)⊗si(~).
Theorem 2.2.9. Then the set
(U(a~), ·, 1, ∆a~, ǫa~
, Sa~, Ra~
)
is a quasi-triangular Hopf QUE algebra and we call it a quantization (Hopf one) of the Lie
bialgebra (a~, r1(~)). We moreover see that it is obtained by a twist via the element
J−1
a~
from the quasi-triangular quasi-Hopf QUE algebra in Subsection 1.1.6.
In particular, we have
Φa~· (Ja~
)12,3 · (Ja~)12 = (Ja~
)1,23 · (Ja~)23,
where the product · is in Ua~.
43
2.2.3 Non-degenerate triangular Lie bialgebras
This is the case (a~, [, ]a~, εa~
= dc(~)r1(~)) where r1(~) is non-degenerate ⇐⇒ invertible , det r1(~) is a unit in IK~ (IK~ is a PID) r1(~) is skew-symmetric
Also, (a~)+ = a~, (a~)− = a~ g~ = a~ ⊕ a~∗
In this case, last theorem is
Theorem 2.2.10. Consider the non-degenerate triangular Lie bialgebra over IK~
(a~, [, ]a~, εa~
= dc(~)r1(~), r1(~) ∈ a~ ⊗ a~, det r1(~) a unit in IK~,
[r1(~), r1(~)]a~= 0),
the set (Ua~, ·, 1, ∆a~, ǫa~
, Sa~, Ra~
) where ∆a~(a~) = J−1
a~· ∆a~
(a~) · Ja~ Sa~(a~) = Q−1 · Sa~
(a~) · Q Ra~= σJ−1
a~· Ja~
, (π ⊗ π)Ω = 0!!!
is a triangular Hopf QUE algebra. We have also the equality
(Ja~)12,3 · (Ja~
)12 = (Ja~)1,23 · (Ja~
)23.
We denote it as
Aa~ ,J
−1a~
meaning that is obtained by a twist via J−1
a~from the usual trivial Hopf triangular algebra
(Ua~, ·, 1,∆a~, ǫa~
, Sa~, Ra~
= 1 ⊗ 1).
Informally, we could say that, in the Hoschild cohomology of Ua~, Ja~is an invariant star
product on the formal ”Lie group” on the ring IK~ whose Lie algebra is the Lie algebra a~
over the ring IK~.
Lemma 2.2.11. Let r1(~) ∈ a~ ⊗IK~a~ as before, that is skew-symmetric and invertible. Let
β1(~) ∈ a∗~∧IK~
a∗~
defined as
β1(~) = (β1(~))abea ⊗ eb
where
r1(~)ab. (β1(~))ac
= δb
c
and then (β1(~))ca.r1(~)ba = δb
c, r1(~)ba. (β1(~))
ca= δb
c. Then,
44
[r1(~), r1(~)]a~= 0 ⇐⇒ dchβ1(~) = 0 µ−1
1 dch µ1(X(~)) ⇐⇒ [r12
1(~), 1 ⊗X(~)]a~
− [X(~) ⊗ 1, r121
(~)]a~, X(~) ∈ a~ Let α(~) ∈ a
∗~. Then
dchα(~) = 0 ⇐⇒ 0 = [r121
(~), 1 ⊗ µ−1
1(α(~))]a~
−
−[µ−1
1(α(~)) ⊗ 1, r12
1(~)]a~
Remember: Poisson coboundary ∂ can be defined in this case as:
−∂ = µ−1
r dch µr
- ∧r+1(a~)
? ?∧
r(a~)
∧r(a~)
∧
r+1(a~)-
µr µr
dch
−∂
and
µ−1
r: ∧r(a~) −→ ∧r(a~)
α 7−→ µ−1
r(α)
where(µ−1
r(α))i1...ir = r
j1i1
1(~) · · · rjrir
1(~)αj1...jr(~)
What we want is to compare the two E-K quantizations we have obtained before of two
different triangular non-degenerate Lie bialgebras defined by different elements r1(~), r′1(~) ∈
a~ ⊗ a~, (a~, [, ]a~, εa~
= dc(~)r1(~)) and (a~, [, ]a~, ε′a~
= dc(~)r′1(~)).
The Lie bracket in a~ is the same for both of them. The Lie bracket of the dual IK~-module
a∗~
is different:
[ξ(~), η(~)]a∗~
= (dc(~)r1(~))t (ξ(~) ⊗ η(~));
[ξ(~), η(~)]′a∗~
=(dc(~)r′1(~)
)t(ξ(~) ⊗ η(~)).
And therefore their doubles:
(a~ ⊕ a∗~, [, ]a~⊕a∗
~, εa~⊕a∗
~= dc(~)r)
(a~ ⊕ a∗~, [, ]
′a~⊕a∗
~
, ε′a~⊕a∗~
= d′c(~)r)
although r is the same in both cases, dc(~) and d′c(~) are different because are defined through
the Lie bracket structures of a~ ⊕ a∗~.
The main point in this comparison is the following classical fact:
45
Theorem 2.2.12. Let G be a Lie group of dimension n and g its Lie algebra. Let Ad be the
adjoint representation of G (on g). Let Ad∧ be the contragradient representation of G (on
g∗), Ad∧ = (Ad g−1)T . Then: Ad∧ induces a representation of G on the exterior algebra ∧g
∗; Ad∧ , g ∈ G, commutes with the Chevalley-Eilenberg diferencial, dch, on g∗; Ad∧ induces a representation, Ad♯, of G on the Chevalley cohomological vector space
H∗ch
(g); The representation Ad♯ is trivial, that is Ad♯g = IdH∗ch
(g), ∀g ∈ G.
We prove a theorem of this type for a Lie algebra a~ over IK~ and mappings Ad(expx(~)) =
exp(ad x(~)) : a~ −→ a~, x(~) ∈ a~.
2.2.4 Interior isomorphisms of (a~, [, ]a~)
Theorem 2.2.13. Let X~ ∈ a~ and let
ϕ1
~ : a~ −→ a~
be defined as
ϕ1
~.Y~ = exp(~ adX~).Y~.
Then ϕ1
~is well defined, that is Im ϕ1
~⊂ a~; ϕ1
~is invertible; ϕ1
~is an isomorphism of Lie algebras.
Our interest is in the contragradient mapping
ϕ2
~ =((ϕ1
~
)t)−1
=(exp
(~ adt
X~
))−1= exp(−~ adt
X~) = exp(~ ad∗
X~)
IK~-module isomorphism of a~. We have
ϕ2
~ ⊗ ϕ2
~ = exp(~ ad∗X~
) ⊗ exp(~ ad∗X~
)
=(
exp(~ ad∗X~
) ⊗ Ida∗~
)
(
Ida∗~⊗ exp(~ ad∗
X~))
= exp(
ad∗~X~⊗ Ida∗
~+ Ida∗
~⊗ ad∗~X~
)
because ad∗~X~
⊗ Ida∗~
and Ida∗~⊗ ad∗
~X~commute.
46
Theorem 2.2.14. Let β~ ∈ a∗~∧ a
∗~
be a Chevalley 2-cocycle, that is,
dchβ~ = 0,
relatively to the Lie algebra structure of a~ and the trivial representation of a~ on IK~.
Then, we have
(i) (ϕ2
~⊗ ϕ2
~)β~ = exp
(
ad∗~X~
⊗ Ida∗~
+ Ida∗~⊗ ad∗
~X~
)
β~
= β~ + dchγ~, where γ~ ∈ a∗~;
(ii) Since β~ is closed, (ϕ2
~⊗ ϕ2
~)β~ is closed;
(iii) If β~ is exact, (ϕ2
~⊗ ϕ2
~)β~ is exact;
(iv) (ϕ2
~⊗ ϕ2
~) acts on H2
ch(a~) and this action is the identity.
Proof. It is enough to prove (i).
We have for n = 1 and any ei, ej ∈ a~ elements in a basis,
⟨(Id⊗ ad∗~X~
+ ad∗~X~⊗ Id) (β~); ei ⊗ ej〉 =
= −〈β~; (Id⊗ ad~X~+ ad~X~
⊗ Id)ei ⊗ ej〉
= −〈β~; ei ⊗ [~X~, ej ]a~+ [~X~, ei]a~
⊗ ej〉
= −~
⟨
β~; ei ⊗Xk
~Cl
kj(~) el +Xk
~Cl
ki(~) el ⊗ ej
⟩
= −~Xk
~Cl
kj(~)(β~)il − ~Xk
~Cl
ki(~)(β~)lj
= −~Xk
~
(
C l
jk(~)(β~)li +C l
ki(~)(β~)lj
)
= −~Xk
~
(
−C l
ij(~) (β~)lk
)
= ~Xk
~ β~ ([ei, ej ]a~, ek)
= ~β~ ([ei, ej ]a~,X~)
= −~β~ (X~, [ei, ej ]a~)
= − (i(~X~)β~) [ei, ej ]a~
= dch (i(~X~)β~) (ei, ej),
where we used that β~ is a 2-cocycle (β~([x, y], z)+β~([y, z], x)+β~([z, x], y) = 0, x, y, z ∈ a~),
Ck
ij(~) are the structure constants of the Lie algebra a~ in a basis ei and dchα(ea ⊗ eb) =
−α([ea, eb]).
So, we obtain
(Id⊗ ad∗~X~+ ad∗~X~
⊗ Id)(β~) = dch(i~X~β~),
for any cocycle β~ and X~ ∈ a~.
47
For any elements ei, ej in a basis of a~, we have (n = 2)
⟨(Id⊗ ad∗~X~
+ ad∗~X~⊗ Id)2β~, ei ⊗ ej
⟩=
=⟨(Id⊗ ad∗~X~
+ ad∗~X~⊗ Id) · dch(i~X~
β~); ei ⊗ ej⟩
= −〈dch(i~X~β~); (Id⊗ ad~X~
+ ad~X~⊗ Id)(ei ⊗ ej)〉
= −〈dch(i~X~β~); ei ⊗ [~X~, ej ]a~
+ [~X~, ei]a~⊗ ej〉
= (i~X~β~)([ei, [~X~, ej ]a~
]a~+ [[~X~, ei]a~
, ej ]a~)
= (i~X~β~)([[ej ,~X~]a~
, ei]a~+ [[~X~, ei]a~
, ej ]a~)
= (i~X~β~)(−[[ei, ej ]a~
,~X~]a~)
= (i~X~β~)([~X~, [ei, ej ]a~
]a~)
= (i~X~β~) ad~X~
([ei, ej ]a~)
= 〈dch(−(i~X~β~) ad~X~
), ei ⊗ ej〉 .
Thus, we get
(Id⊗ ad∗~X~+ ad∗~X~
⊗ Id)2(β~) = −dch ((i(~X~)β~) ad~X~) .
For n = 3, we have
⟨(Id⊗ ad∗~X~
+ ad∗~X~⊗ Id)3β~, ea ⊗ eb
⟩=
=⟨−(Id⊗ ad∗~X~
+ ad∗~X~⊗ Id) dch ((i~X~
β~) ad(~X~)) ; ea ⊗ eb⟩
= 〈dch ((i~X~β~) ad(~X~)) ; (Id⊗ ad~X~
+ ad~X~⊗ Id)(ea ⊗ eb)〉
= 〈dch((i~X~β~) ad(~X~)) ; ea ⊗ [~X~, eb]a~
+ [~X~, ea]a~⊗ eb〉
= −〈(i~X~β~) ad(~X~) ; [ea, [~X~, eb]a~
]a~+ [[~X~, ea]a~
, eb]a~〉
= −〈(i~X~β~) ad(~X~); [[eb,~X~]a~
, ea]a~+ [[~X~, ea]a~
, eb]a~〉
= 〈(i~X~β~) ad(~X~); [[ea, eb]a~
,~X~]a~〉
= −〈(i~X~β~) ad(~X~) ad(~X~); [ea, eb]a~
〉
= 〈dch ((i~X~β~) ad(~X~) ad(~X~)) ; ea ⊗ eb〉 .
Then,
(Id⊗ ad∗~X~+ ad∗~X~
⊗ Id)3β~ = dch ((i~X~β~) ad(~X~) ad(~X~)) .
We have obtained
(ϕ2
~ ⊗ ϕ2
~)β~ = β~ + ~ dch(i(X~)β~)+
+ ~2dch
(
−1
2!(i(X~)β~) adX~
)
+
+ ~3dch
(1
3!(i(X~)β~) adX~
adX~
)
+ . . .
48
Let us write
γ1(~) = (i(X~)β~) ∈ a∗~
γ2(~) =
(
−1
2!(i(X~)β~) adX~
)
∈ a∗~
γ3(~) =
(1
3!(i(X~)β~) adX~
adX~
)
∈ a∗~.
Then, we have
(ϕ2
~ ⊗ ϕ2
~)β~ = β~ + dch(~γ1(~) + ~2γ2(~) + ~
3γ3(~) + . . . )
and it is easy to get a general formula for γk(~) ∈ a∗~, k ∈ IN.
Let us compute the first terms in powers of ~ of γ1(~), γ2(~), γ3(~), etc.
γ1(~) = i(X~)β~ = i(Xa ~a−1)(βb ~
b−1) = (i(Xa)βb) ~a+b−2
=∑
R≥0
(∑
a+b=R+2
i(Xa)βb
)
~R
= i(X1)β1 + [i(X2)β1 + i(X1)β2] ~+
+ [i(X1)β3 + i(X2)β2 + i(X3)β1] ~2+
+ [ . . . ]~3 + . . . ,
where Xa ∈ a, βb ∈ a∗ ⊗IK a
∗ and therefore i(Xa)βb ∈ a∗.
γ2(~) = −1
2!((i(X~)β~) adX~
) = −1
2!
(
(i(Xm ~m−1)(βl ~
l−1) adXp ~p−1
)
= −1
2!
((i(Xm)βl) adXp
)~
m+l+p−3.
As Xm ∈ a, βl ∈ a∗ ⊗IK a
∗, the map i(Xm)βl sends a to IK, that is, i(Xm)βl ∈ a∗. But
relatively to adXp , even being Xp ∈ a, this map doesn’t send a in a but a into a~.
Let ei be a basis of (a, IK), then
adXpei = [Xp, ei]a~= [X l
pel, ei]a~
= X l
p[el, ei]a~
= X l
pCk
li(~)ek
= X l
p
(
Ck
li(s)~s−1
)
ek =(
X l
pCk
li(s)ek
)
~s−1 (2.2)
where we have used
Ck
li(~) =
∑
s≥1
Ck
li(s)~s−1, Ck
li(s) ∈ a.
Lemma 2.2.15. Let us consider the following mappings
Bs : a × a −→ a
(y, z) 7−→ Bs(y, z)
49
where
Bs(y, z) = Bs(ylel, z
iei) = ylziBs(el, ei) = ylziCk
li(s)ek.
Then, we have
(i) The definition of Bs is independent of the basis ei on (a, IK);
(ii) Bs is a IK-bilinear mapping on a with values in a;
(iii) [el, ei]a~= Bs(el, ei)~
s−1.
We can write (2.2) as
adXpei = (X l
pCk
li(s)ek)~s−1
= X l
pBs(el, ei)~
s−1 = Bs(Xl
pel, ei)~
s−1 = Bs(Xp, ei)~s−1
= ((i(Xp)Bs)ei)~s−1 = ((i(Xp)Bs)~
s−1)ei.
Therefore
adXp = (i(Xp)Bs) ~s−1, s = 1, 2, 3, . . .
Here i(Xp)Bs ∈ HomIK(a, a) and doesn’t contain ~.
Returning to the expression of γ2(~), we can write now
γ2(~) = −1
2!
((i(Xm)βl) adXp
)~
m+l+p−3
= −1
2!
((i(Xm)βl)
((i(Xp)Bs)~
s−1))
~m+l+p−3
= −1
2!((i(Xm)βl) (i(Xp)Bs)) ~
m+l+p+s−4
=∑
R≥0
∑
m+l+p+s=R+4
m,l,p,s≥1
−1
2!((i(Xm)βl) (i(Xp)Bs))
~R
= −1
2![(i(X1)β1) (i(X1)B1)] +
−1
2![(i(X2)β1) (i(X1)B1) + (i(X1)β2) (i(X1)B1)+
+(i(X1)β1) (i(X2)B1) + (i(X1)β1) (i(X1)B2)] ~+
+ [. . . ] ~2 + . . .
50
We have also
γ3(~) =1
3!((i(X~)β~) adX~
adX~)
=1
3!
(
i(Xp~p−1)(βa~
a−1) (adXb~
b−1) (adXq~q−1)
)
=1
3!((i(Xp)βa) (i(Xb)Bs) (i(Xq)Br)) ~
p+a+b+q+s+r−6
=∑
M≥0
∑
p+a+b+q+s+r=M+6
p,a,b,q,s,r≥1
1
3!((i(Xp)βa) (i(Xb)Bs) (i(Xq)Br))
~M
=1
3![(i(X1)β1) (i(X1)B1) (i(X1)B1)]+
+1
3![(i(X2)β1) (i(X1)B1) (i(X1)B1)+
+ (i(X1)β2) (i(X1)B1) (i(X1)B1)+
+ (i(X1)β1) (i(X2)B1) (i(X1)B1)+
+ (i(X1)β1) (i(X1)B2) (i(X1)B1)
+ (i(X1)β1) (i(X1)B1) (i(X2)B1)+
+(i(X1)β1) (i(X1)B1) (i(X1)B2)] ~+
+ [. . . ] ~2 + . . .
Then we get
(ϕ2
~ ⊗ ϕ2
~)β~ = β~ + dch(~γ1(~) + ~2γ2(~) + ~
3γ3(~) + . . . )
= β~ + dch [i(X1)β1] ~+
+ dch
[
i(X2)β1 + i(X1)β2 −1
2!(i(X1)β1) (i(X1)β1)
]
~2
+ dch [i(X1)β3 + i(X2)β2 + i(X3)β1+
−1
2!(i(X2)β1 i(X1)B1 + i(X1)β2 i(X1)B1+
+ i(X1)β1 i(X2)B1 + i(X1)β1 i(X1)B2)+
+1
3!i(X1)β1 i(X1)B1 i(X1)B1
]
~3 + . . .
51
Let us define the following elements in a∗,
α1 = i(X1)β1
α2 = i(X2)β1 + i(X1)β2 −1
2!(i(X1)β1) (i(X1)β1 (2.3)
α3 = i(X1)β3 + i(X2)β2 + i(X3)β1 −1
2!(i(X2)β1 i(X1)B1+
+ i(X1)β2 i(X1)B1 + i(X1)β1 i(X2)B1 + i(X1)β1 i(X1)B2)+
+1
3!i(X1)β1 i(X1)B1 i(X1)B1
...
αk =k∑
j=1
k−j+1∑
i=1
(−1)i+1
i!
∑
a1+a2+b2···+ai+bi=i+k−j
a1,a2,b2,...,ai,bi≥1
(
(iXa1βj) iXa2
Bb2· · · iXai
Bbi
)
then
(ϕ2
~ ⊗ ϕ2
~)β~ = β~ +∑
k≥1
dch(αk)~k, αk ∈ a∗.
We then prove a converse of last theorem.
Theorem 2.2.16. Let
α(~) = αk~k ∈ a
∗~, αk ∈ a
∗, k = 1, 2, 3, . . . ,
there exists a unique element
X~ = Xl~l−1, Xl ∈ a
verifying the equality
(exp(ad∗~X~
) ⊗ exp(ad∗~X~))β~ = β~ + dchα(~).
Proof. If X~ = Xl~l−1 exists verifying last equality, it must be obtained from the equalities
(2.3).
Due to the invertibility of β1, from the first equality we determine X1. Then, because we
know at this step X1, β1, β2, α2 and β1 is invertible, from the second equality we determine
X2. X3 can be computed from the third equality because β1 is invertible and at this step we
know X1, X2, B1, B2, β1, β2 and β3, etc, etc..
Recall (a~, [, ]a~) is a finite-dimensional Lie algebra over IK~ which is a deformation Lie algebra
of the Lie algebra (a, [, ]a) over IK.
52
r1(~), r′1(~) are two elements of a~ ⊗IK~
a~ which are non degenerate skew-symmetric
and solutions of YBE relatively to the Lie algebra (a~, [, ]a~) and such that the terms in
power ~0 coincide and are equal to r1 ∈ a ⊗IK a. (a~, [, ]a~, r1(~)), (a~, [, ]a~
, r′1(~)) are the corresponding finite dimensional Lie bialgebras
over IK~, both deformation Lie bialgebras for the Lie bialgebra (a, [, ]a, r1) over IK. Let ϕ1
~: a~ −→ a~ be a Lie algebra automorphism of (a~, [, ]a~
). Let ψ~ : a∗r1(~)
−→ a∗r′1(~)
be a Lie algebra isomorphism from the Lie algebra (a∗r1(~)
, [, ]a∗r1(~)
)
to the Lie algebra (a∗r′1(~), [, ]a∗
r′1(~)
). Let (ϕ1
~, ψ~) : a~ ⊕ a
∗r1(~)
−→ a~ ⊕ a∗r′1(~)
be a Lie algebra isomorphism between the
doubles. Let ϕ1
~: U(a~) −→ U(a~) and ψ~ : U(a∗
r1(~)) −→ U(a∗
r′1(~)
) be the (which are continuous
in the (~)-adic topology) associative algebra isomorphisms extensions respectively of ϕ1
~
and ψ~.
Theorem 2.2.17. Let Jr1(~)
a~and J
r′1(~)
a~be invariant star products obtained in Theorem 3.10.
Suppose β1(~) ∈ a∗~∧IK~
a∗~
is defined as β1(~) = (β1(~))abea ⊗ eb where r1(~)ab. (β1(~))
ac= δb
c
and β′1(~) ∈ a
∗~∧IK~
a∗~
is defined in a similar way from r′1(~).
Then, Jr1(~)
a~and J
r′1(~)
a~are equivalent star products if, and only if, β1(~) and β′
1(~) belong
to the same cohomological class. In other words, Jr1(~)
a~and J
r′1(~)
a~are equivalent star products
if, and only if, there exists a 1-cochain α(~) ∈ a∗~
such that
β′1(~) = β1(~) + dchα(~).
Sketch of the proof (⇐=) [β1(~)] = [β′1(~)], there exists X~ ∈ a~ such that
exp(ad∗~X~)⊗
2
β1(~) = β′1(~). Then ϕ1
~= exp(ad~X~
) is a Lie algebra isomorphism a~ −→ a~ and (ϕ1
~⊗ ϕ1
~)r1(~) =
r′1(~). The map (ϕ1
~, ((ϕ1
~)t)−1) is a Lie bialgebra isomorphism between a~ ⊕ a
∗r1(~)
and a~ ⊕
a∗r′1(~)
. We prove that Jr′1~
a~= (ϕ1
~⊗ ϕ1
~)Jr1~
a~. Using a theorem by Drinfeld, there exists an element u = exp(~X~) such that
Jr′1~
a~= ∆a(u) · J
r1~
a~· (u−1 ⊗ u−1)
and u−1 defines the equivalence.
53
Sketch of the proof (=⇒) Hochschid cohomology on U(a~) appears. We construct an element α(~) ∈ a∗~
such that
β′1(~) = β1(~) + dchα(~)
where α(~) = α1~ + α2~2 + . . . .
2.2.5 Hochschild cohomology on the coalgebra Ua~
From a theorem by Cartier, we get
H∗Hoch
(Γ(a~)) = ∧(a~)
where Γ(a~) is the coalgebra of divided powers (see also Bourbaki). But from Cartier and
Bourbaki, we know
Γ(a~) ≃ TS(a~) ≃ S(a~)
as IK~ bialgebras (making a proof similar to the classical one for Lie algebras over a field of
characteristic 0)
We have also an isomorphism
S(a~) ≃ U(a~)
as coalgebras. From these isomorphisms, we get what we want
H∗Hoch
(U(a~)) ≃ ∧(a~).
Acknowledgements
The second author has been partially supported by CMUC and FCT- project POCI/MAT/58452/2004.
References
[1] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, and D. Sternheimer, Deformation
theory and quantization, Ann. Phys. 111 (1978), 61 – 151.
[2] P. Cartier, Hyperalgebres et Groups de Lie formels, Seminaire Sophus Lie, 2eme anne,
Faculte des Sciences de Paris, 19.
[3] V.G. Drinfeld, On constant, quasiclassical solutions of the Yang-Baxter equations, Sov.
Math. Dokl. 28 (1983), 667 – 671.
[4] V.G. Drinfeld, Quasi–Hopf algebras, Leningrad Math. J. 1 (6) (1990), 1419 – 1457.
54
[5] V.G. Drinfeld, On quasitriangular quasi–Hopf algebras and a group closely connected with
Gal(Q/Q), Leningrad Math. J. 2 (4) (1991), 829 – 860.
[6] D. Kazhdan and G. Lusztig, Tensor structures arising from affine Lie algebras, III, J.
Amer. Math. Soc. 7 (1994), 335 – 381.
[7] P. Etingof and D. Kazhdan, Quantization of Lie Bialgebras, I, Selecta Mathematica 2 (1)
(1996), 1 – 41.
[8] M. Gerstenhaber, On the deformation of rings and algebras, Ann. of Math. 79 (1964), 59
– 103.
[9] S. Majid, Foundations of Quantum Group Theory, Cambridge University Press, 1995.
[10] C. Moreno and L. Valero, Star products and quantization of Poisson-Lie groups, J. Geom.
Phys. 9 (4) (1992), 369 – 402.
[11] C. Moreno and L. Valero, Star products and quantum groups, in Physics and Manifolds,
Proceedings of the International Colloquium in honour of Yvonne Choquet-Bruhat, Paris,
June 3-5 (1992), Mathematical Physics Studies, vol. 15, Kluwer Academic Publishers, 1994,
pp. 203–234.
[12] C. Moreno and L. Valero, The set of equivalent classes of invariant star products on
(G;β1), J. Geom. Phys. 23 (4) (1997), 360 – 378.
[13] C. Moreno and L. Valero, in Y.Choquet-Bruhat and C. De-Witt-Morette,Morette, Anal-
ysis, Manifolds and Physics. Part II, North-Holland, Amsterdam, 2000.
[14] C. Moreno and J. Teles, Preferred quantizations of nondegenerate triangular Lie bialge-
bras and Drinfeld associators, Research and Expositions in Mathematics, Volume 25 (2002)
347-362, 2002 Heldermann Verlag Berlin.
[15] C. Moreno and J. Teles, A classification Theorem of Invariant Star Products in the
setting of Etingof-Kazhdan quantization,Proceedings of the XII Fall Workshop on Geometry
and Physics, Coimbra, 2003. Publicaciones de la RSME, vol 7, pp. 223-228.
[16] N.Yu. Reshetikhin and M.A. Semenov-Tian-Shansky, Quantum R-matrices and factor-
ization problems, J. Geom. Phys. 5 (4) (1988), 533 – 550.
[17] J. Teles, The formal Chevalley cohomology in the quantization by Etingof and Kazh-
dan of non-degenerate triangular Lie bialgebras and their classical doubles, Ph.D. thesis,
Departamento de Matematica, Universidade de Coimbra, 2003.
55
56
On central extension of Leibniz n–algebras
J. M. Casas∗
Abstract
The study of central extensions of Leibniz n-algebras by means of homological methods
is the main goal of the paper. Thus induced abelian extensions are introduced and the
classification of various classes of central extensions depending on the character of the
homomorphism θ∗(E) in the five-term exact sequence
nHL1(K) → nHL1(L)θ∗(E)→ M → nHL0(K) → nHL0(L) → 0
associated to the abelian extension E : 0 → Mκ→ K
π→ L → 0 is done. Homological char-
acterizations of this various classes of central extensions are given. The universal central
extension corresponding to a perfect Leibniz n-algebra is constructed and characterized.
The endofunctor uce which assigns to a perfect Leibniz n-algebra its universal central
extension is described. Functorial properties are obtained and several results related with
the classification in isogeny classes are achieved. Finally, for a covering f : L′ ։ L (a
central extension with L′ a perfect Leibniz n-algebra), the conditions under which an
automorphism or a derivation of L can be lifted to an automorphism or a derivation of
L′ are obtained.
1 Introduction
The state of a classical dynamic system is described in Hamiltonian mechanics by means
of N coordinates q1, . . . , qN and N momenta p1, . . . , pN . The 2N variables q1, . . . , pN are
referred as canonical variables of the system. Other physically important quantities as energy
and momentum are functions F = F (q, p) of the canonical variables. These functions, called
observables, form an infinite dimensional Lie algebra with respect to the Poisson bracket
F,G =∑
N
i=1( ∂F
∂qi
∂G
∂pi− ∂F
∂pi
∂G
∂qi). The equation of motion are qi = ∂H
∂piand pi = −∂H
∂qi, where
the Hamiltonian operator of the system H is the total energy. These equations may be written
in terms of Poisson brackets as qi = qi,H; pi = pi,H. In general the time evolution of an
observable F is given by F = F,H.
The simplest phase space for Hamiltonian mechanics is R2 with coordinates x, y and
canonical Poisson bracket f1, f2 = ∂f1
∂x
∂f2
∂y− ∂f1
∂y
∂f2
∂x= ∂(f1,f2)
∂(x,y). This bracket satisfies the
∗Dpto. Matematica Aplicada I, Universidad de Vigo, E. U. I. T. Forestal, Campus Universitario A Xun-
queira, 36005 Pontevedra, Spain. Email:[email protected]
57
Jacobi identity f1, f2, f3+f3, f1, f2+f2, f3, f1 = 0 and give rise to the Hamilton
equations of motion df
dt= H, f.
In 1973, Nambu [23] proposed the generalization of last example defining for a tern of
classical observables on the three dimensional space R3 with coordinates x, y, z the canonical
bracket given by f1, f2, f3 = ∂(f1,f2,f3)
∂(x,y,z)where the right hand side is the Jacobian of the
application f = (f1, f2, f3) : R3 → R
3. This formula naturally generalizes the usual Poisson
bracket from a binary to a ternary operation on the classical observables. The Nambu-
Hamilton generalized motion equations include two Hamiltonian operators H1 and H2 and
have the form df
dt= H1,H2, f. For the canonical Nambu bracket the following fundamental
identity holds
f1, f2, f3, f4, f5 + f3, f1, f2, f4, f5 + f3, f4, f1, f2, f5 =
f1, f2, f3, f4, f5
This formula can be considered as the most natural generalization, at least from the dynamical
viewpoint, of the Jacobi identity. Within the framework of Nambu mechanics, the evolution
of a physical system is determined by n−1 functionsH1, . . . ,Hn−1 ∈ C∞(M) and the equation
of motion of an observable f ∈ C∞(M) is given by df/dt = H1, . . . ,Hn−1, f.
These ideas inspired novel mathematical structures by extending the binary Lie bracket
to a n-ary bracket. The study of this kind of structures and its application in different areas
as Geometry and Mathematical Physics is the subject of a lot of papers, for example see [8],
[10], [11], [12], [13], [14], [16], [22], [24], [25], [26], [27], [28], [29], [30] and references given
there.
The aim of this paper is to continue with the development of the Leibniz n-algebras theory.
Concretely, using homological machinery developed in [3], [4], [5], [8] we board an extensive
study of central extensions of Leibniz n-algebras. Thus in Section 3 we deal with induced
abelian extensions and Section 4 is devoted to the classification of various classes of central
extensions depending on the character of the homomorphism θ∗[E] in the five-term exact
sequence
nHL1(K) → nHL1(L)θ∗(E)
→ M → nHL0(K) → nHL0(L) → 0
associated to the abelian extension E : 0 → Mκ→ K
π→ L → 0. Homological characterizations
of this various classes of central extensions are given. When we restrict to the case n = 2 we
recover results on central extensions of Leibniz algebras in [2], [6], [7], [9].
Sections 5 and 6 are devoted to the construction and characterization of universal central
extensions of perfect Leibniz n-algebras. We construct an endofunctor uce which assigns to a
perfect Leibniz n-algebra its universal central extension. Functorial properties are obtained
and several results related with the classification in isogeny classes are achieved. Finally, in
Section 7 we analyze the conditions to lift an automorphism or a derivation of L to L′ in a
covering (central extension where L′ is a perfect Leibniz n-algebra) f : L ։ L′.
58
2 Preliminaries on Leibniz n-algebras
A Leibniz n-algebra is a K-vector space L equipped with a n-linear bracket [−, . . . ,−] : L⊗n →
L satisfying the following fundamental identity
[[x1, x2, . . . , xn], y1, y2, . . . , yn−1] =
n∑
i=1
[x1, . . . , xi−1, [xi, y1, y2, . . . , yn−1], xi+1, . . . , xn] (1)
A morphism of Leibniz n-algebras is a linear map preserving the n-bracket. Thus we have
defined the category of Leibniz n-algebras, denoted by nLeib. In case n = 2 the identity (1)
is the Leibniz identity, so a Leibniz 2-algebra is a Leibniz algebra [18, 19, 20], and we use
Leib instead of 2Leib.
Leibniz (n + 1)-algebras and Leibniz algebras are related by means of the Daletskii’s
functor [10] which assigns to a Leibniz (n + 1)-algebra L the Leibniz algebra Dn(L) = L⊗n
with bracket
[a1 ⊗ · · · ⊗ an, b1 ⊗ · · · ⊗ bn] :=n∑
i=1
a1 ⊗ · · · ⊗ [ai, b1, . . . , bn] ⊗ · · · ⊗ an (2)
Conversely, if L is a Leibniz algebra, then also it is a Leibniz n-algebra under the following
n-bracket [8]
[x1, x2, . . . , xn] := [x1, [x2, . . . , [xn−1, xn]]] (3)
Examples:
1. Examples of Leibniz algebras in [1], [19] provides examples of Leibniz n-algebras with
the bracket defined by equation (3).
2. A Lie triple system [17] is a vector space equipped with a ternary bracket [−,−,−] that
satisfies the same identity (1) (particular case n = 3) and, instead of skew-symmetry,
satisfies the conditions [x, y, z] + [y, z, x] + [z, x, y] = 0 and [x, y, y] = 0. It is an easy
exercise to verify that Lie triple systems are non-Lie Leibniz 3-algebras.
3. Rn+1 is a Leibniz n-algebra with the bracket given by [
→x1,
→x2, . . . ,
→xn] := det(A), where
A is the following matrix
→e1
→e2 . . .
→en+1
x11 x21 . . . x(n+1)1
x12 x22 . . . x(n+1)2
. . . . . . . . . . . .
x1n x2n . . . x(n+1)n
Here→xi= x1i
→e1 +x2i
→e2 + · · ·+x(n+1)i
→en+1 and
→e1,
→e2, . . . ,
→en+1 is the canonical basis
of Rn+1.
59
4. An associative trialgebra is a K-vector space A equipped with three binary operations:
⊣,⊥,⊢ (called left, middle and right, respectively), satisfying eleven associative relations
[21]. Then A can be endowed with a structure of Leibniz 3-algebra with respect to the
bracket
[x, y, z] = x ⊣ (y ⊥ z) − (y ⊥ z) ⊢ x− x ⊣ (z ⊥ y) + (z ⊥ y) ⊢ x
= x ⊣ (y ⊥ z − z ⊥ y) − (y ⊥ z − z ⊥ y) ⊢ x
for all x, y, z ∈ A.
5. Let C∞(Rn) be the algebra of C∞-functions on Rn and x1, . . . , xn be the coordinates on
Rn. Then C∞(Rn) equipped with the bracket [f1, . . . , fn] = det( ∂fi
∂xj)i,j=1,...,n is a n-Lie
algebra [12], so also it is a Leibniz n-algebra.
Let L be a Leibniz n-algebra. A subalgebra K of L is called n-sided ideal if [l1, l2, . . . , ln] ∈
K as soon as li ∈ K and l1, . . . , li−1, li+1, . . . , ln ∈ L, for all i = 1, 2, . . . , n. This definition
guarantees that the quotient L/K is endowed with a well defined bracket induced naturally
by the bracket in L.
Let M and P be n-sided ideals of a Leibniz n-algebra L. The commutator ideal of
M and P, denoted by [M,P,Ln−2], is the n-sided ideal of L spanned by the brackets
[l1, . . . , li, . . . , lj , . . . , ln] as soon as li ∈ M, lj ∈ P and lk ∈ L for all k 6= i, k 6= j; i, j, k ∈
1, 2, . . . , n. Obviously [M,P,Ln−2] ⊂ M∩P. In the particular case M = P = L we obtain
the definition of derived algebra of a Leibniz n-algebra L. If L = [L, n. . .,L] = [Ln], then the
Leibniz n-algebra is called perfect.
For a Leibniz n-algebra L, we define its centre as the n-sided ideal
Z(L) = l ∈ L | [l1, . . . , li−1, l, li+1, . . . , ln] = 0,∀li ∈ L, i = 1, . . . , i, . . . , n
An abelian Leibniz n-algebra is a Leibniz n-algebra with trivial bracket, that is, the
commutator n-sided ideal [Ln] = [L, . . . ,L] = 0. It is clear that a Leibniz n-algebra L is
abelian if and only if L = Z(L). To any Leibniz n-algebra L we can associate its largest
abelian quotient Lab. It is easy to verify that Lab∼= L/[Ln].
A representation of a Leibniz n-algebra L is a K-vector space M equipped with n actions
[−, . . . ,−] : L⊗i⊗ M ⊗L⊗(n−1−i) → M, 0 ≤ i ≤ n − 1, satisfying (2n − 1) axioms which are
obtained from (1) by letting exactly one of the variables x1, . . . , xn, y1, . . . , yn−1 be in M and
all the others in L.
If we define the multilinear applications ρi : L⊗n−1 → EndK(M) by
ρi(l1, . . . , ln−1)(m) = [l1, . . . , li−1,m, li, . . . , ln−1], 1 ≤ i ≤ n− 1
ρn(l1, . . . , ln−1)(m) = [l1, . . . , ln−1,m]
then the axioms of representation can be expressed by means of the following identities [3]:
60
1. For 2 ≤ k ≤ n,
ρk([l1, . . . , ln], ln+1, . . . , l2n−2) =n∑
i=1
ρi(l1, . . . , li, . . . , ln) · ρk(li, ln+1, . . . , l2n−2)
2. For 1 ≤ k ≤ n,
[ρ1(ln, . . . , l2n−2), ρk(l1, . . . , ln−1)] =
n−1∑
i=1
ρk(l1, . . . , li−1, [li, ln, . . . , l2n−2], li+1, . . . , ln−1)
being the bracket on EndK(M) the usual one for associative algebras.
A particular instance of representation is the case M =L, where the applications ρi are
the adjoint representations
adi(l1, . . . , ln−1)(l) = [l1, . . . , li−1, l, li, . . . , ln−1], 1 ≤ i ≤ n− 1
adn(l1, . . . , ln−1)(l) = [l1, . . . , ln−1, l]
If the components of the representation ad : L⊗n−1 → EndK(L) are ad = (ad1, . . . , adn),
then Ker ad = l ∈ L | adi(l1, . . . , ln−1)(l) = 0,∀(l1, . . . , ln−1) ∈ L⊗n−1, 1 ≤ i ≤ n, that is,
Ker ad is the centre of L.
Now we briefly recall the (co)homology theory for Leibniz n-algebras developed in [3, 8].
Let L be a Leibniz n-algebra and let M be a representation of L. Then Hom(L,M) is a
Dn−1(L)-representation as Leibniz algebras [8]. One defines the cochain complex nCL∗(L,M)
to be CL∗(Dn−1(L),Hom(L,M)). We also put nHL∗(L,M) := H∗(nCL
∗(L,M)). Thus,
by definition nHL∗(L,M) ∼= HL∗(Dn−1(L),Hom(L,M)). Here CL⋆ denotes the Leibniz
complex and HL⋆ its homology, called Leibniz cohomology (see [19, 20] for more information).
In case n = 2, this cohomology theory gives 2HLm(L,M) ∼= HLm+1(L,M), m ≥ 1 and
2HL0(L,M) ∼= Der(L,M). On the other hand, nHL
0(L,M) ∼= Der(L,M) and nHL1(L,M)
∼= Ext(L,M), where Ext(L,M) denotes the set of isomorphism classes of abelian extensions
of L by M [8].
Homology with trivial coefficients of a Leibniz n-algebra L is defined in [3] as the homology
of the Leibniz complex nCL⋆(L) := CL⋆(Dn−1(L),L), where L is endowed with a structure
of Dn−1(L) symmetric corepresentation [20]. We denote the homology groups of this complex
by nHL⋆(L). When L is a Leibniz 2-algebra, that is a Leibniz algebra, then we have that
2HLk(L) ∼= HLk+1(L), k ≥ 1. Particularly, 2HL0(L) ∼= HL1(L) ∼= L/[L,L] = Lab. On the
other hand, nHL0(L) = Lab and nHL1(L) ∼= (R ∩ [Fn])/[R,Fn−1] for a free presentation
0 → R → F → L → 0.
Moreover, to a short exact sequence 0 → M → K → L → 0 of Leibniz n-algebras we can
associate the following five-term natural exact sequences [3]:
0 → nHL0(L, A) → nHL
0(K, A) → HomL(M/[M,M,Kn−2], A) →
61
nHL1(L, A) → nHL
1(K, A) (4)
for every L-representation A, and
nHL1(K) → nHL1(L) → M/[M,Kn−1] → nHL0(K) → nHL0(L) → 0 (5)
3 Induced abelian extensions
An abelian extension of Leibniz n-algebras is an exact sequence E : 0 → Mκ→ K
π→ L → 0
of Leibniz n-algebras such that [k1, . . . , kn] = 0 as soon as ki ∈ M and kj ∈ M for some
1 ≤ i, j ≤ n (i. e., [M,M,Kn−2] = 0). Here k1, . . . , kn ∈ K. Clearly then M is an abelian
Leibniz n-algebra. Let us observe that the converse is true only for n = 2. Two such extensions
E and E′ are isomorphic when there exists a Leibniz n-algebra homomorphism from K to K′
which is compatible with the identity on M and on L. One denotes by Ext(L,M) the set of
isomorphism classes of extensions of L by M .
If E is an abelian extension of Leibniz n-algebras, then M is equipped with a L-representation
structure given by
[l1, . . . , li−1,m, li+1, . . . , ln] = [k1, . . . , ki−1, κ(m), ki+1, . . . , kn]
such that π(kj) = lj , j = 1, . . . , i− 1, i+ 1, . . . , n, i = 1, 2, . . . , n.
The abelian extensions of Leibniz n-algebras are the objects of a category whose mor-
phisms are the commutative diagrams of the form:
0 // M1
α
κ1 // K1
β
π1 // L1
γ
// 0
0 // M2
κ2 // K2
π2 // L2// 0
We denote such morphism as (α, β, γ) : (E1) → (E2). It is evident that α and γ satisfy the
following identities
α([l1, . . . , li−1,m, li+1, . . . , ln]) = [γ(l1), . . . , γ(li−1), α(m), γ(li+1), . . . , γ(ln)]
i = 1, 2, . . . , n, provided than M2 is considered as L1-representation via γ. That is, α is a
morphism of L1-representations.
Given an abelian extension E and a homomorphism of Leibniz n-algebras γ : L1 → L we
obtain by pulling back along γ an extension Eγ of M by L1, where Kγ = K ×L L1, together
with a morphism of extensions (1, γ′, γ) : Eγ → E. We call to the the extension (Eγ) the
backward induced extension of E.
Proposition 1. Every morphism (α, β, γ) : E1 → E of abelian extensions of Leibniz n-
algebras admits a unique factorization of the form
E1
(α,η,1)
→ Eγ
(1,γ′,γ)
→ E
62
Given a homomorphism of L-representations α : M → M0 we obtain the extension αE :
0 → M0
κ0→ αKπ0→ L → 0 by putting αK = (M0⋊K)/S, where S = (α(m),−κ(m)) | m ∈M.
We call to the the extension αE the forward induced extension of E.
Proposition 2. Every morphism (α, β, γ) : E → E0 of abelian extensions of Leibniz n-
algebras admits a unique factorization of the form
E(α,α
′,1)
→ (αE)(1,ξ,γ)
→ (E0)
through the forward induced extension determined by α.
4 Various classes of central extensions
Let E : 0 → Mκ→ K
π→ L → 0 ∈ Ext(L,M) be. Since M is a L-representation, we have
associated to it the exact sequence (4)
0 → Der(L,M)Der(π)
→ Der(K,M)ρ
→ HomL(M,M)θ∗(E)
→ nHL1(L,M)
π∗
→ nHL1(K,M)
Then we define ∆ : Ext(L,M) → nHL1(L,M),∆([E]) = θ∗(E)(1M). The naturality of the
sequence (4) implies the well definition of ∆. Now we fix a free presentation 0 → Rχ
→ Fǫ→
L → 0, then there exists a homomorphism f : F → K such that π.f = ǫ, which restricts to
f : R → M. Moreover f induces a L-representation homomorphism ϕ : R/[R,R,Fn−2] → M
where the action from L on R/[R,R,Fn−2] is given via ǫ, that is,
[l1, . . . , li−1, r, li+1, . . . , ln] = [x1, . . . , xi−1, r, xi+1, . . . , xn] + [R,R,Fn−2]
where ǫ(xj) = lj , j ∈ 1, . . . , i− 1, i+ 1, . . . , n, i ∈ 1, . . . , n. The naturality of sequence (4)
induces the following commutative diagram
Der(K,M)
f∗
// HomL(M,M)
ϕ∗
θ∗(E)
//nHL
1(L,M) //nHL
1(K,M)
f∗
Der(F ,M)τ∗
// HomL(R/[R,R,Fn−2],M)σ∗
//nHL
1(L,M) //nHL
1(F ,M)
Having in mind that nHL1(F ,M) = 0 [8], then ∆[E] = θ∗(E)(1M) = σ∗ϕ∗(1M) = σ∗(ϕ).
Proposition 3. ∆ : Ext(L,M) → nHL1(L,M) is an isomorphism.
Proof. It is a tedious but straightforward adaptation of the Theorem 3.3, p 207 in [15]. ⋄
Definition 1. Let E : 0 →Mκ→ K
π→ L → 0 be an extension of Leibniz n-algebras. We call
E central if [M,Kn−1] = 0.
63
Associated to E we have the isomorphism nHLk(L,M)
θ∗∼= Hom(nHLk(L),M) (see The-
orem 3 in [3]). Let us observe that M is a trivial L-representation since E is a central
extension. On the other hand, ∆[E] ∈ nHL1(L,M), then θ∗∆[E] ∈ Hom(nHL1(L),M).
Moreover θ∗∆[E] = θ∗(E), being θ∗(E) the homomorphism given by the exact sequence (5):
nHL1(K) → nHL1(L)θ∗(E)
→ M → nHL0(K) → nHL0(L) → 0
When 0 → M → K → L → 0 is a central extension, the sequence (5) can be enlarged
with a new term as follows (see [5]):
⊕n−1
i=1Ji → nHL1(K) → nHL1(L) → M → nHL0(K) → nHL0(L) → 0 (6)
where Ji = (M⊗ n−i. . . ⊗M ⊗ Kab⊗ i. . . ⊗Kab) ⊕ (M⊗ n−i−1. . . ⊗M ⊗ Kab ⊗ M ⊗ Kab⊗ i−1. . .
⊗Kab) ⊕ · · · ⊕ (Kab⊗ i. . . ⊗Kab ⊗ M⊗ n−i. . . ⊗M).
According to the character of the homomorphism θ∗∆[E] we can classify the central
extensions of Leibniz n-algebras. Thus we have the following
Definition 2. The central extension E : 0 →Mκ→ K
π→ L → 0 is called:
1. Commutator extension if θ∗∆[E] = 0.
2. Quasi-commutator extension if θ∗∆[E] is a monomorphism.
3. Stem extension if θ∗∆[E] is an epimorphism.
4. Stem cover if θ∗∆[E] is an isomorphism.
Let us observe that Definition 2 in case n = 2 agrees with the definitions in [2] for Leibniz
algebras. It is clear, by naturality of sequence (5), that the property of a central extension
which belongs to any of the described classes only depends on the isomorphism class.
Following, we characterize the various classes defined in terms of homological properties.
Proposition 4. The following statements are equivalent:
1. E is a commutator extension.
2. 0 → M → nHL0(K) → nHL0(L) → 0 is exact.
3. π∗ : [Kn]∼→ [Ln].
4. M ∩ [Kn] = 0.
64
Proof. E is a commutator extension ⇔ θ∗(E) = 0 ⇔ 0 → M → nHL0(K) → nHL0(L) →
0 is exact (use sequence (5)) ⇔ M ∩ [Kn] = 0 ⇔ [Kn]∼→ [Ln] (use next diagram)
M ∩ [Kn]
// // [Kn]
π∗ // // [Ln]
M
// // K
π // // L
M
M∩[Kn]// // Kab
π // // Lab
(7)
⋄
Proposition 5. The following statements are equivalent:
1. E is a quasi-commutator extension.
2. nHL1(π) : nHL1(K) → nHL1(L) is the zero map.
3. 0 → nHL1(L) → M → nHL0(K) → nHL0(L) → 0 is exact.
4. nHL1(L) ∼= M ∩ [Kn].
Proof. θ∗(E) is a monomorphism ⇔ Ker θ∗(E) = Im π∗ = 0 ⇔ π∗ : nHL1(K) → nHL1(L)
is the zero map ⇔ 0 → nHL1(L) → M → nHL0(K) → nHL0(L) → 0 is exact (by exactness
in sequence (5)).
For the equivalence of last statement we must use that the monomorphism θ∗(E) factors
as nHL1(L) ։ M ∩ [Kn] → M. ⋄
Corollary 1. If E is a quasi-commutator extension with nHL1(L) = 0, then E is a commu-
tator extension.
Proof. M ∩ [Kn] ∼= nHL1(L) = 0. ⋄
Corollary 2. Let E be a central extension with K a free Leibniz n-algebra, then E is a
quasi-commutator extension.
Proof. nHL1(K) = 0 [8] and use sequence (5). ⋄
Proposition 6. The following statements are equivalent:
1. E is a stem extension.
2. κ∗ : M → nHL0(K) is the zero map.
3. π : nHL0(K)∼→ nHL0(L).
4. M ⊆ [Kn].
65
Proof. In exact sequence (5), θ∗(E) is an epimorphism ⇔ κ∗ = 0 ⇔ π∗ is an isomorphism.
In diagram (7), Kab∼= Lab ⇔
M
M∩[Kn]= 0 ⇔ M ⊆ [Kn]. ⋄
Proposition 7. Every class of central extensions of a L-trivial representation M is forward
induced from a stem extension.
Proof. Pick any central extension class E : 0 → M → K → L → 0, then θ⋆(E) :
nHL1(L) → M factors as i.τ : nHL1(L) ։ M ∩ [Kn] = M1 → M. As M1 is a trivial L-
representation, then given τ there exists a central extension E1 ∈ nHL1(L,M1) such that
θ⋆(E1) = τ . Moreover, E1 is a stem extension since θ⋆(E1) is an epimorphism. By naturality
of sequence (5) on the forward construction E1 → i(E1), we have that θ⋆i(E1) = iθ∗(E1) =
iτ = θ⋆(E), i. e., i(E1) = E, and so E is forward induced by E1, which is a stem extension. ⋄
Proposition 8. Let L be a Leibniz n-algebra and let U be a subspace of nHL1(L), then there
exists a stem extension E with U = Ker θ⋆∆[E].
Proof. We consider the quotient vector space M = nHL1(L)/U as a L-trivial represen-
tation. We consider the central extension E : 0 → M → K → L → 0 ∈ nHL1(L,M). Thus
θ⋆∆[E] = θ⋆(E) ∈ Hom(nHL1(L),M). If θ⋆(E) : nHL1(L) → M = nHL
1(L)/U is the canon-
ical projection, then there exists a central extension E : 0 → M → K → L → 0 such that
θ⋆∆[E] = θ⋆(E) is the canonical projection. Associated to E we have the exact sequence (5),
in which U = Ker θ⋆(E) = Ker θ⋆∆[E]. Moreover E is a stem extension since θ⋆∆[E] = θ⋆(E)
is an epimorphism. ⋄
Proposition 9. The following statements are equivalent:
1. E is a stem cover.
2. π : nHL0(K)∼→ nHL0(L) and nHL1(π) : nHL1(K) → nHL1(L) is the zero map.
Proof. θ⋆(E) is an isomorphism in sequence (6). ⋄
Corollary 3. A stem extension is a stem cover if and only if U = 0.
Proof. U = 0 ⇔ Ker θ⋆(E) = 0 ⇔ θ⋆(E) : nHL1(L) → M is an isomorphism. ⋄
5 Universal central extensions
Definition 3. A central extension E : 0 → Mκ→ K
π→ L → 0 is called universal if for
every central extension E′ : 0 → M ′ → K′ π′
→ L → 0 there exists a unique homomorphism
h : K → K′ such that π′h = π.
Lemma 1. Let 0 → N → Hπ→ L → 0 be a central extension of Leibniz n-algebras, being
H a perfect Leibniz n-algebra. Let 0 → M → Kσ→ L → 0 be another central extension of
Leibniz n-algebras. If there exists a homomorphism of Leibniz n-algebras φ : H → K such
that σφ = π, then φ is unique.
66
Proof. See Lemma 5 in [4]. ⋄
Lemma 2. If 0 → N → Hρ
→ K → 0 and 0 → M → Kπ→ L → 0 are central extensions with
K a perfect Leibniz n-algebra, then 0 → L = Ker π ρ→ Hπρ→ L → 0 is a central extension.
Moreover, if K is a universal central extension of L, then 0 → N → Hρ
→ K → 0 splits.
Proof. Since K is a perfect Leibniz n-algebra, then ρ restricts to the epimorphism ρ′ :
Z(H) ։ Z(K). From this argument and using classical techniques it is an easy task to end
the proof. ⋄
Lemma 3. If K is a perfect Leibniz n-algebra and π : K ։ L is an epimorphism, then L is
a perfect Leibniz n-algebra.
Lemma 4. If E : 0 → Mi→ K
π→ L → 0 is a universal central extension, then K and L are
perfect Leibniz n-algebra.
Proof. Assume that K is not a perfect Leibniz n-algebra, then Kab is an abelian Leibniz
n-algebra and, consequently, is a trivial L-representation. We consider the central exten-
sion E : 0 → Kab → Kab × Lpr
→ L → 0, then the homomorphisms of Leibniz n-algebras
ϕ,ψ : K → Kab × L, ϕ(k) = (k, π(k));ψ(k) = (0, π(k)), k ∈ K verify that pr ϕ = π = pr ψ,
so E can not be a universal central extension. Lemma 3 ends the proof. ⋄
Theorem 1.
1. If E : 0 → M → Kπ→ L → 0 is a central extension with K a perfect Leibniz n-algebra
and every central extension of K splits, if and only if E is universal.
2. A Leibniz n-algebra L admits a universal central extension if and only if L is perfect.
3. The kernel of the universal central extension is canonically isomorphic to nHL1(L,K).
Proof. See Theorem 5 in [3]. The equivalence of statement (1) is due to Lemma 2. ⋄
Corollary 4. The central extension 0 → M → Kπ→ L → 0 is universal if and only if
nHL0(K) = nHL1(K) = 0 (that is, K is a superpefect Leibniz n-algebra).
Proof. By Theorem 1 (1) K is a perfect Leibniz n-algebra, so nHL0(K) = Kab = 0. On
the other hand, the splitting of any central extension by K is equivalent to nHL1(K) = 0,
since 0 → 0 → K∼→ K → 0 is the universal central extension of K (that is, K is centrally
closed) . ⋄
Corollary 5. Let E : 0 → M → Kπ→ L → 0 be a central extension with L a perfect Leibniz
n-algebra. E is a stem cover if and only if nHL0(K) = nHL1(K) = 0.
67
Proof. From the exact sequence (5) associated to E and Proposition 9. ⋄
Remark. Let us observe that the stem cover of a perfect Leibniz n-algebra L is isomorphic
to the universal central extension of L.
Proposition 10. Let E : 0 → M → Kπ→ L → 0 be a central extension of Leibniz n-algebras.
If K is a perfect Leibniz n-algebra, then the sequence 0 → nHL1(K) → nHL1(L) → M → 0
is exact and nHL0(L) = 0.
Proof. L is perfect by Lemma 3. Exact sequence (6) associated to E ends the proof. ⋄
Following we achieve a construction of the universal central extension of a perfect Leibniz
n-algebra slightly different to the construction given in [4]. This approach permits us to obtain
new results concerning to the endofunctor uce which assigns to a perfect Leibniz n-algebra its
universal central extension. To do this, we recall that the computation of the homology with
trivial coefficients of a Leibniz n-algebra L (see [3]) uses the chain complex
· · · → nCL2(L) = CL2(L⊗n−1,L) = L⊗2n−1 d2→ nCL1(L) = CL1(L
⊗n−1,L) = L⊗n d1→
d1→ nCL0(L) = CL0(L⊗n−1,L) = L → 0
where
d2(x1 ⊗ · · · ⊗ x2n−1) = [x1, x2, . . . , xn] ⊗ xn+1 ⊗ · · · ⊗ x2n−1−
n∑
i=1
x1 ⊗ · · · ⊗ xi−1 ⊗ [xi, xn+1, . . . , x2n−1] ⊗ xi+1 ⊗ · · · ⊗ xn
and
d1(x1 ⊗ · · · ⊗ xn) = [x1, x2, . . . , xn]
Let L be a perfect Leibniz n-algebra. As K-vector spaces, we consider the submodule I of
L⊗n spanned by the elements of the form
[x1, x2, . . . , xn] ⊗ xn+1 ⊗ · · · ⊗ x2n−1−
n∑
i=1
x1 ⊗ · · · ⊗ xi−1 ⊗ [xi, xn+1, . . . , x2n−1] ⊗ xi+1 ⊗ · · · ⊗ xn
for all x1, . . . , x2n−1 ∈ L. Let us observe that I = Im d2. Then we construct uce(L) = L⊗n/I.
We denote by x1, . . . , xn the element x1 ⊗ · · · ⊗ xn + I of uce(L). By construction the
following identity holds in uce(L)
[x1, x2, . . . , xn], xn+1, . . . , x2n−1 =
n∑
i=1
x1, . . . , xi−1, [xi, xn+1, . . . , x2n−1], xi+1, . . . , xn (8)
68
The linear map d1 vanishes on the elements of I, so it induces an epimorphism d : uce(L) → L
which is defined by d(x1, x2, . . . , xn) = [x1, x2, . . . , xn]. Let us observe that Ker d = Ker
d1/I = Ker d1/Im d2 = nHL1(L). On the other hand, the K-vector space uce(L) is endowed
with a structure of Leibniz n-algebra by means of the following n-ary bracket:
[x1, x2, . . . , xn] := d(x1), d(x2), . . . , d(xn)
for all x1, . . . , xn ∈ uce(L). In this way d becomes into a Leibniz n-algebras homomorphism.
Particularly, the following identity holds:
[x11, . . . , x1n, x21, . . . , x2n, . . . , xn1, . . . , xnn] =
[x11, . . . , x1n], [x21, . . . , x2n], . . . , [xn1, . . . , xnn]
Consequently, d : uce(L) ։ L is an epimorphism of Leibniz n-algebras. Actually d : uce(L) ։
[Ln], but if L is perfect, then L = [Ln]. It is an easy task, using identity (8) and Lemma 1,
to verify that the epimorphism d : uce(L) ։ [Ln] is the universal central extension of L when
L is a perfect Leibniz n-algebra.
By the uniqueness and having in mind the constructions of the universal central extension
given in [4] we derive that uce(L) ∼= [Fn]/[R,Fn−1] ∼= L⋆ n. . . ⋆L, where 0 → R → F → L → 0
is a free presentation of the perfect Leibniz n-algebra L and ⋆ denotes a non-abelian tensor
product of Leibniz n-algebras introduced in [4].
As we can observe, last construction does not depend on the perfectness of the Leibniz
n-algebra L, that is, in general case, we have constructed the universal central extension of
[Ln]. Following we explore the functorial properties of this construction. So we consider a
homomorphism of perfect Leibniz n-algebras f : L′ → L. Let IL′ and IL be as the submod-
ule defined previously. The canonical application f⊗n : nCL1(L′) = L′⊗n → nCL1(L) =
L⊗n, f⊗n(x1 ⊗ · · · ⊗ xn) = f(x1) ⊗ · · · ⊗ f(xn) maps IL′ into IL, thus it induces a linear
map uce(f) : uce(L′) → uce(L), uce(f)x1, . . . , xn = f(x1), . . . , f(xn). Moreover uce(f) is
a homomorphism of Leibniz n-algebras.
On the other hand, one verifies by construction that the following diagram is commutative:
uce(L′)uce(f)
//
d′
uce(L)
d
L′f
// L
(9)
Thus we have a right exact covariant functor uce : nLeib → nLeib and, consequently, an
automorphism f of L gives rise to an automorphism uce(f) of uce(L). The commutativity
of last diagram implies that uce(f) leaves nHL1(L) invariant. Thus, we obtain the group
homomorphism
Aut(L) → g ∈ Aut(uce(L)) : g(nHL1(L)) = nHL1(L) : f 7→ uce(f)
69
6 Isogeny classes in nLeib
Definition 4. A Leibniz n-algebra L is said to be unicentral if every central extension π :
K → L maps Z(K) onto Z(L):
Z(K) // //
Z(L)
Ker(π)
::
::v
vv
vv
vv
vv
// // Kπ // // L
Proposition 11. Let L be a perfect Leibniz n-algebra, then L is unicentral.
Proof. Lab = 0, so the Ganea map C : ⊕n
i=1Ji → nHL1(L) given by the exact sequence
(6) associated to the central extension 0 → Z(L) → L → L/ZL → 0 is the zero map, so
Corollary 3.6 in [5] ends the proof. ⋄
Under certain hypothesis the composition of universal central extensions is again a uni-
versal central extension:
Corollary 6. Let 0 → N → Hτ→ K → 0 and 0 → M → K
π→ L → 0 be two central
extensions of Leibniz n-algebras. Then π τ : H ։ L is a universal central extension if and
only if τ : H ։ K is a universal central extension.
Proof. If π τ : H ։ L is a universal central extension, then nHL0(H) = nHL1(H) = 0
by Corollary 4, this Corollary also implies that 0 → N → Hτ→ K → 0 is a universal central
extension.
Conversely, if 0 → N → Hτ→ K → 0 is a universal central extension, then K is perfect
by Theorem 1, (2), then π τ : H ։ L is a central extension by Lemma 2. Now Corollary 4
ends the proof. ⋄
Proposition 12. For every perfect Leibniz n-algebra L there is the isomorphism LZ(L)
∼=uce(L)
Z(uce(L)).
Proof. If L is perfect, then L is unicentral by Proposition 11, so in the following diagram
Z(uce(L)) // //
Z(L)
uce(L) // // L
the kernels of the horizontal morphisms coincide, and then the cokernels of the vertical mor-
phisms are isomorphic. ⋄
Remark: For a Leibniz n-algebra L satisfying that L/Z(L) is unicentral, then the central
extension 0 → Z(L) → L → L/Z(L) → 0 implies that Z(L/Z(L)) = 0 (see Corollary 3.8 in
[5]). Hence for a perfect Leibniz n-algebra L, L/Z(L) is perfect, so it is unicentral and by
Proposition 12 we have that LZ(L)
∼=uce(L)
Z(uce(L))are centerless, that is its center is trivial.
70
Corollary 7. Let L and L′ be two perfect Leibniz n-algebras. Then uce(L) ∼= uce(L′) if and
only if L/Z(L) ∼= L′/Z(L′).
Proof. If uce(L) ∼= uce(L′), then, by Proposition 12, L/Z(L) ∼= uce(L)/Z(uce(L)) ∼=
uce(L′)/Z(uce(L′)) ∼= L′/Z(L′).
Conversely, if L/Z(L) ∼= L′/Z(L′) then uce(L/Z(L)) ∼= uce(L′/Z(L′)). Corollary 6 applied
to the central extensions uce(L) ։ L and L ։ L/Z(L) yields the isomorphism uce(L/Z(L)) ∼=
uce(L), from which the result is derived. ⋄
Definition 5. We say that the perfect Leibniz n-algebras L and L′ are isogenous if uce(L) ∼=
uce(L′).
Remarks:
i) Isogeny classes are in an obvious bijection with centerless perfect Leibniz n-algebras: by
Corollary 7 two different centerless perfect Leibniz n-algebras have different universal
central extensions and they are non-isogenous, and in each isogeny class, namely the
class of a perfect Leibniz n-algebra L there is always an isogenous centerless perfect
Leibniz n-algebra L/Z(L).
ii) Isogeny classes are also in bijection with superperfect Leibniz n-algebras: in each isogeny
class, namely the class of a perfect Leibniz n-algebra L there is always an isogenous
superperfect Leibniz n-algebra uce(L), and two different superperfect Leibniz n-algebras
have different universal central extensions and are non-isogenous since a superperfect
Leibniz n-algebra L is centrally closed.
iii) Remark the fact that each isogeny class C is the set of central factors of its superperfect
representant K, and then C is an ordered set. In this ordered set there is a maximal
element, which is the superperfect representant, and a minimal element, which is the
centerless representant.
iv) A Leibniz n-algebra L is called capable if there exists a Leibniz n-algebra K such that
L ∼= K/Z(K). Capable Leibniz n-algebras are characterized in [5]. From Proposition 12
and Corollaries 3.2 and 3.8 in [5] we derive that centerless perfect Leibniz n-algebras
are equivalent to capable perfect Leibniz n-algebras.
The following holds for the centerless representant L and the superperfect representant
uce(L) of each isogeny class:
Corollary 8. Let L be a centerless perfect Leibniz n-algebra. Then Z(uce(L)) = nHL1(L)
and the universal central extension of L is 0 → Z(uce(L)) → uce(L) → L → 0
Proof. Since L is centerless then Proposition 12 implies that 0 → Z(uce(L)) → uce(L) →
L → 0 is isomorphic to the universal central extension of L. ⋄
71
7 Lifting automorphisms and derivations
Lifting automorphisms: Let f : L′ ։ L be a covering (that means a central extension
with L′ a perfect Leibniz n-algebra). Consequently, L also is a perfect Leibniz n-algebra by
Lemma 3, so diagram (9) becomes into
nHL1(L′)
nHL1(L)
uce(L′)
uce(f)//
d′
uce(L)
d
L′f
// // L
(10)
By Corollary 6, the central extension f.d′ : uce(L′) ։ L is universal. Moreover uce(f)
is a homomorphism from this universal central extension to the universal central extension
d : uce(L) → L. Consequently uce(f) is an isomorphism (two universal central extensions of
L are isomorphic). So we obtain a covering d′.uce(f)−1 : uce(L) ։ L′ with kernel
C := Ker (d′.uce(f)−1) = uce(f)(Ker d′) = uce(f)(nHL1(L′))
Theorem 2. (lifting of automorphisms) Let f : L′ ։ L be a covering.
a) Let be h ∈ Aut(L). Then there exists h′ ∈ Aut(L′) such that the following diagram
commutes
L′
h′
f// // L
h
L′
f// // L
(11)
if and only if the automorphism uce(h) of uce(L) satisfies uce(h)(C) = C.
In this case, h′ is uniquely determined by (11) and h′(Ker f) = Ker f .
b) With the notation in statement a), the map h 7→ h′ is a group isomorphism
h ∈ Aut(L) : uce(h)(C) = C → g ∈ Aut(L′) : g(Ker f) = Ker(f)
Proof. a) If h′ exists, then it is a homomorphism from the covering h.f to the covering f ,
so h′ is unique by Lemma 1. By applying the functor uce(−) to the diagram (11) we obtain
the following commutative diagram:
uce(L′)
uce(h′)
uce(f)// uce(L)
uce(h)
uce(L′)
uce(f)// // uce(L)
72
Hence
uce(h)(C) = uce(h).uce(f)(nHL1(L′)) =
uce(f).uce(h′)(nHL1(L′)) = uce(f)(nHL1(L
′)) = C
Conversely, from diagram (10) one derives that d = f.d′.uce(f)−1, and hence one obtains
the following diagram:
C // //
uce(L)
uce(h)
d′.uce(f)−1
// L′
h′
f// // L
h
C // // uce(L)
d′.uce(f)−1
// L′f
// // L
If uce(h)(C) = C, then d′.uce(f)−1.uce(h)(C) = d′.uce(f)−1(C) = 0, then there exists a
unique h′ : L′ → L′ such that h′.d′.uce(f)−1 = d′.uce(f)−1.uce(h). On the other hand,
h.f.d′.uce(f)−1 = f.d′.uce(f)−1.uce(h) = f.h′.d′.uce(f)−1, so h.f = f.h′.
Commutativity of (10) implies that h′(Kerf) = Kerf .
b) By a), the map is well-defined. It is a monomorphism by uniqueness in (a) and it is
an epimorphism, since every g ∈ Aut(L′) with g(Kerf) = Kerf induces an automorphism
h : L → L such that h.f = f.g. Hence, by a), g = h′ and uce(h)(C) = C. ⋄
Corollary 9. If L is a perfect Leibniz n-algebra, then the map
Aut(L) → g ∈ Aut(uce(L)) : g(nHL1(L)) = nHL1(L) : f 7→ uce(f)
is a group isomorphism. In particular, Aut(L) ∼= Aut(uce(L)) if L is centerless.
Proof. We apply statement b) in Theorem 2 to the covering d : uce(L) ։ L. In this case
C = 0 and uce(f)(0) = 0.
If A is centerless, then nHL1(L) = Z(uce(L)) by Corollary 8. Since any automorphism
leaves the center invariant, then the second claim is a consequence of the first one. ⋄
Lifting derivations: Let L be a Leibniz n-algebra and d ∈ Der(L). The linear map
ϕ : L⊗n → L⊗n, ϕ(x1⊗· · ·⊗xn) =∑
n
i=1x1⊗· · ·⊗xi−1⊗d(xi)⊗xi+1⊗· · ·⊗xn, leaves invariant
the submodule I and hence it induces a linear map uce(d) : uce(L) −→ uce(L), X1, . . . ,Xn 7→
d(X1),X2, . . . ,Xn+ X1, d(X2), . . . ,Xn+ · · ·+ X1,X2, . . . , d(Xn) which commutes the
following diagram:
uce(L)
d
uce(d)// uce(L)
d
Ld
// L
In particular, uce(d) leaves Ker (d) invariant. Moreover, a tedious but straightforward
verification stands that uce(d) is a derivations of uce(L). On the other hand,
uce : Der(L) → δ ∈ Der(uce(L)) : δ(nHL1(L)) ⊆ nHL1(L) : d 7→ uce(d)
73
is a homomorphism of Lie algebras (Der(L) is a Lie algebra, see p. 193 in [8]). Its kernel is
contained in the subalgebra of those derivations vanishing on [Ln]. It is also verified that
uce(ad[x11,...,xn1]⊗···⊗[x1,n−1,...,xn,n−1]) = adx11,...,xn1⊗···⊗x1,n−1,...,xn,n−1
where adX1⊗···⊗Xn−1: L → L,X 7→ [X,X1, . . . ,Xn−1] and
adx11,...,xn1⊗···⊗x1,n−1,...,xn,n−1 : uce(L) → uce(L)
z1, . . . , zn 7→ [z1, . . . , zn, x11, . . . , xn1, . . . ,
x1,n−1, . . . , xn,n−1].
.
Hence uce(add(X1)⊗···⊗d(Xn−1)
) = adX1⊗···⊗Xn−1, being Xi = xi1, . . . , xin, i = 1, . . . , n − 1,
and uce(ad[Ln]⊗n−1
... ⊗[Ln]) = IDer(uce(L)), where IDer(uce(L)) are the inner derivations
dX1⊗···⊗Xn−1: uce(L) → uce(L), z1, . . . , zn 7→ [z1, . . . , zn,X1, . . . ,Xn−1],
with Xi = xi1, . . . , xin ∈ uce(L), i = 1, . . . , n − 1.
The functorial properties of the functor uce(−) concerning derivations are described in the
following
Lemma 5. Let f : L′ → L be a homomorphism of Leibniz n-algebras, let d ∈ Der(L) and
d′ ∈ Der(L′) be such that f.d′ = d.f , then uce(f).uce(d′) = uce(d).uce(f)
Proof. A straightforward computation on the typical elements x1, . . . , xn of uce(L)
shows the commutativity. ⋄
Theorem 3. (lifting of derivations) Let f : L′ ։ L be a covering. We denote C =
uce(f)(nHL1(L′)) j nHL1(L).
a) A derivation d of L lifts to a derivation d′ of L′ satisfying d.f = f.d′ if and only if the
derivation uce(d) of uce(L) satisfies uce(d)(C) j C. In this case, d′ is uniquely determined
and leaves Ker f invariant.
b) The map
d ∈ Der(L) : uce(d)(C) j C → δ ∈ Der(L′) : δ(Ker(f)) j Ker(f)
d 7−→ d′
is an isomorphism of Lie algebras mapping IDer(L) onto IDer(L′).
c) For the covering d : uce(L) ։ L, the map
uce : Der(L) → δ ∈ Der(uce(L)) : δ(nHL1(L)) j nHL1(L)
is an isomorphism preserving inner derivations. If L is centerless, then Der(L) ∼= Der(uce(L)).
74
Proof. a) Assume the existence of a derivation d′ such that f.d′ = d.f , then by Lemma 5
we have that
uce(d)(C) = uce(d).uce(f)(nHL1(L′)) =
uce(f).uce(d′)(nHL1(L′)) j uce(f)(nHL1(L
′)) = C
The converse is parallel to the proof of Theorem 2 a) having in mind that d is a derivation if
and only if Id+ d is an automorphism.
b) The K-vector space Der(L) is endowed with a structure of Lie algebra by means of the
bracket [d1, d2] = d1.d2 − d2.d1. Now the proof easily follows from a).
c) Apply b) to the covering d : uce(L) ։ L. Observe that C = 0 in this situation.
If L is centerless, then by Corollary 8 we have that nHL1(L) = Z(uce(L)) and a derivation
of L leaves the center invariant, so the isomorphism follows from b). ⋄
Acknowledgments
Supported by MCYT, Grant BFM2003-04686-C02-02 (European FEDER support included)
and Xunta de Galicia, Grant PGIDIT04PXIC37101PN.
References
[1] S. A. Ayupov and B. A. Omirov, On Leibniz algebras, Algebra and Operator Theory,
Tashkent, (1997), 1–12, Kluwer Acad. Publ., Dordrecht (1998).
[2] J. M. Casas, Central extensions of Leibniz algebras, Extracta Mathematicae, Vol. 13
(1998), N. 3, 393–397.
[3] J. M. Casas, Homology with trivial coefficients of Leibniz n-algebras, Comm. in Algebra
31 (3) (2003), 1377–1386.
[4] J. M. Casas, A non-abelian tensor product and universal central extension of Leibniz n-
algebras, Bull. Belgian Math. Soc Simon Stevin 10 (2004), 259–270.
[5] J. M. Casas, Ganea term for homology of Leibniz n-algebras, Algebra Colloquium 12 (4)
(2005), 629–634.
[6] J. M. Casas, E. Faro and A. M. Vieites, Abelian extensions of Leibniz algebras, Comm. in
Algebra 27 (6) (1999), 2833–2846.
[7] J. M. Casas and M. Ladra, Stem extensions and stem covers of Leibniz algebras, Georgian
Math. J. 9 (4) (2002), 659–669.
[8] J. M. Casas, J.-L. Loday and T. Pirashvili, Leibniz n-algebras, Forum Math 14 (2) (2002),
189–207.
75
[9] J. M. Casas and A. M. Vieites, Central extensions of perfect of Leibniz algebras, Research
and Exposition in Mathematics, Vol. 25 (2002), 189–196.
[10] Y. I. Daletskii and L. A. Takhtajan, Leibniz and Lie algebras structures for Nambu
algebra, Letters in Math. Physics 39 (1997), 127–141.
[11] A. S. Dzhumadil’daev, Representations of vector product n-Lie algebras, Comm. in Al-
gebra 32 (9) (2004), 3315–3326.
[12] V. T. Filippov, n-Lie algebras, Sib. Mat. Zh., Vol. 26 (1985), No. 6, 126-140.
[13] P. Gautheron, Some remarks concerning Nambu mechanics, Letters in Math. Physics 37
(1996), 103–116.
[14] J. Grabowski and G. Marmo, On Filippov algebroids and multiplicative Nambu-Poisson
structures, Differential Geom. Appl. 12 (2000), no. 1, 35–50.
[15] P. J. Hilton and U. Stammbach, A course in homological algebra, Graduate Text in Math.
4 (1971), Springer-Verlag, New York-Berlin.
[16] R. Ibanez, M. de Leon, B. Lopez, J. C. Marrero and E. Padron, Duality and modular
class of a Nambu-Poisson structure, J. Phys. A: Math. Gen. 34 (2001), 3623–3650.
[17] W. Lister, A structure theory of Lie triple systems, Trans. Amer. Math. Soc. 72 (1952),
217–242.
[18] J.-L. Loday, Cyclic homology, Grundl. Math. Wiss. Bd. 301 (1992), Springer.
[19] J.-L. Loday, Une version non commutative des algebres de Lie: les algebres de Leibniz,
L’Enseignement Mathematique 39 (1993), 269–292.
[20] J.-L. Loday and T. Pirashvili, Universal enveloping algebras of Leibniz algebras and
(co)homology, Math. Ann. 296 (1993), 139–158.
[21] J.-L. Loday and M. O. Ronco, Trialgebras and families of polytopes, Homotopy theory:
relations with algebraic geometry, group cohomology and algebraic K-theory, Contemp.
Math. 346 (2004), Am. Math. Soc., Providence, RI, 369–398.
[22] G. Marmo, G. Vilasi and A. M. Vinogradov, The local structure of n-Poisson and n-
Jacobi manifolds, J. Geom. Phys. 25 (1998), 141–182.
[23] Y. Nambu, Generalized Hamiltonian dynamics, Phys. Rev. D 3 (7) (1973), 2405–2412.
[24] A. P. Pojidaev, Enveloping algebras of Filippov algebras, Comm. in Algebra 31 (2) (2003),
883–900.
[25] P. Saraiva, Reduced n-Lie algebras, Comm. in Algebra 30 (4) (2002), 2057–2074.
76
[26] J. Stasheff, The (secret?) homological algebra of the Batalin-Vilkovisky approach. In “Sec-
ondary calculus and cohomological physics” (Moscow, 1997), 195–210, Contemp. Math.,
219 (1998), Amer. Math. Soc., Providence, RI.
[27] L. A. Takhtajan, On fundation of the generalized NAMBU mechanics, Comm. Math,
Phys. 160 (1994), No. 2, 295–315.
[28] L. A. Takhtajan, A higher order analog of the Chevalley-Eilenberg complex and the de-
formation theory of n-gebras, St. Petersburg Math. J. Vol. 6 (1995), No. 2, 429–438.
[29] I. Vaisman, A survey on Nambu-Poisson brackets, Acta Math. Univ. Comenian. (W. S.)
68 (1999), no. 2, 213–241.
[30] A. Vinogradov and M. Vinogradov, On multiple generalizations of Lie algebras and Pois-
son manifolds. In “Secondary calculus and cohomological physics” (Moscow, 1997), 273–
287, Contemp. Math., 219 (1998), Amer. Math. Soc., Providence, RI.
77
78
Fine gradings on some exceptional algebras
Cristina Draper∗ Candido Martın†
Abstract
We describe the fine gradings, up to equivalence, on the exceptional Lie algebras of
least dimensions, g2 and f4.
1 Introduction
The research activity around gradings on Lie algebras has grown in the last years. Many
works on the subject could be mentioned but, for briefness, we shall cite [2] and [5]. The
most known fine grading on a simple Lie algebra, that is, the decomposition in root spaces,
has shown to have many applications to the Lie algebras theory and to representation theory.
So, it seems that other fine gradings could give light about different aspects of these algebras.
This paper is based in [3] and [4]. In the first one we classify up to equivalence all
the gradings on g2, and, in the second one all the nontoral gradings on f4. In this last
algebra we rule out the study of the toral gradings because they provide essentially the
same perspective of the algebra than the root space decomposition. The gradings which
summarize the information about all the ways in which an algebra can be divided are the fine
gradings, because any grading is obtained by joining homogeneous spaces of a fine grading.
Our purpose in these pages is provide a complete description of them, in the cases of g2 and
f4. This objective does not need so technical tools as those used in the above works. These
technicalities are only needed to show that, effectively, the described gradings will cover all
the possible cases.
Besides we will describe the fine gradings on the Cayley algebra and on the Albert algebra,
motivated by their close relationship to g2 and f4 respectively.
2 About gradings and automorphisms
Let F be an algebraically closed field of characteristic zero, which will be used all through this
work. If V is an F -algebra and G an abelian group, we shall say that the decomposition V =
∗Departamento de Matematica Aplicada, Campus de El Ejido, S/N, 29071 Malaga, Spain.†Departamento de Algebra, Geometrıa y Topologıa, Campus de Teatinos, S/N. Facultad de Ciencias, Ap.
59, 29080 Malaga, Spain. Supported by the Spanish MCYT projects MTM2004-06580-C02-02 and MTM2004-
08115-C04-04, and by the Junta de Andalucıa PAI projects FQM-336 and FQM-1215
79
⊕g∈GVg is a G-grading on V whenever for all g, h ∈ G, VgVh ⊂ Vgh. The set g ∈ G | Vg 6= 0
is called the support of the grading and denoted by Supp(G). We shall always suppose that G
is generated by Supp(G). We say that two gradings V = ⊕g∈GXg = ⊕h∈HYh are equivalent
if the sets of homogeneous subspaces are the same up to isomorphism, that is, there are an
automorphism f ∈ aut(V ) and a bijection between the supports α : Supp(G) → Supp(H)
such that f(Xg) = Yα(g) for any g ∈ Supp(G). A convenient invariant for equivalence is
that of type. Suppose we have a grading on a finite-dimensional algebra, then for each
positive integer i we will denote by hi the number of homogeneous components of dimension
i. Besides we shall say that the grading is of type (h1, h2, . . . , hl), for l the greatest index
such that hl 6= 0. Of course the number∑
iihi agrees with the dimension of the algebra.
There is a close relationship between group gradings and automorphisms. More precisely,
if f1, . . . , fn ⊂ aut(V ) is a set of commuting semisimple automorphisms, the simultaneous
diagonalization becomes a group grading, and conversely, given V = ⊕g∈GVg a G-grading, the
set of automorphisms of V such that every Vg is contained in some eigenspace is an abelian
group formed by semisimple automorphisms.
Consider an F -algebra V , a G-grading V = ⊕g∈GXg and an H-grading V = ⊕h∈HYh.
We shall say that the H-grading is a coarsening of the G-grading if and only if each nonzero
homogeneous component Yh with h ∈ H is a direct sum of some homogeneous components
Xg. In this case we shall also say that the G-grading is a refinement of the H-grading. A
group grading is fine if its unique refinement is the given grading. In such a case the group
of automorphisms above mentioned is a maximal abelian subgroup of semisimple elements,
usually called a MAD (”maximal abelian diagonalizable”). It is convenient to observe that
the number of conjugacy classes of MADs groups of aut(V ) agrees with the number of equiv-
alence classes of fine gradings on V . Our objective is to describe the fine gradings, up to
equivalence, on the exceptional Lie algebras g2 and f4.
3 Gradings on C and Der(C)
Under the hypotesis about the ground field there is only one isomorphy class of Cayley
algebras so that we take one and forever any representative C of the class. Consider also
g2 := Der(C). The Lie algebra g2 is generated by the set of derivations Dx,y | x, y ∈ C,
where
Dx,y = [lx, ly] + [lx, ry] + [rx, ry] ∈ Der(C)
for lx and rx the left and right multiplication operators in C.
In our context a grading on an algebra is always induced by a set of commuting diagonal-
izable automorphisms of the algebra. Thus, an important tool for translating gradings on C
80
to gradings on g2 (and conversely) is given by the isomorphism of algebraic groups
Ad: aut(C) → aut(g2)
f 7→ Ad(f); Ad(f)(d) := fdf−1.
Since Ad(f)(Dx,y) = Df(x),f(y), the grading induced on g2 by the grading C = ⊕g∈GCg
is given by g2 = L = ⊕g∈GLg with Lg =∑
g1+g2=gDCg1
,Cg2. As pointed out in [3] this
translation procedure has the drawback that equivalent gradings on C are not necessarily
transformed into equivalent gradings on g2. But of course isomorphic gradings are indeed
transformed into isomorphic ones and reciprocally. Moreover two fine equivalent gradings on
C are transformed into fine equivalent gradings on g2 and reciprocally.
3.1 Gradings on the Cayley algebra
Next we fix a basis of C given by:
B = (e1, e2, u1, u2, u3, v1, v2, v3).
This is called the standard basis of the Cayley algebra C, and is defined for instance in [3,
Section 3] by the following relations
e1uj = uj = uje2,
e2vj = vj = vje1,
uiuj = vk = −ujui,
−vivj = uk = vjvi,
uivi = e1,
viui = e2,
where e1 and e2 are orthogonal idempotents, (i, j, k) is any cyclic permutation of (1, 2, 3), and
the remaining relations are null. This algebra is isomorphic to the Zorn matrices algebra. For
further reference we recall here that the standard involution x 7→ x of C is the one permuting
e1 and e2 and making x = −x for x = ui or vj (i, j = 1, 2, 3). This enable us to define the
norm n : C → F by n(x) := xx, the trace map tr : C → F by tr(x) := x+ x, and the subspace
C0 of all trace zero elements. This is f -invariant for any f ∈ aut(C).
The gradings on C are computed in [6] in a more general context. In particular, up to
equivalence, there are only two fine gradings on C. These are:
a) The Z2-toral grading given by
C0,0 = 〈e1, e2〉
C0,1 = 〈u1〉 C1,1 = 〈u2〉 C−1,−2 = 〈u3〉
C0,−1 = 〈v1〉 C−1,−1 = 〈v2〉 C1,2 = 〈v3〉
whose homogeneous elements out of the zero component C0,0 have zero norm.
b) The Z3
2-nontoral grading given by
C000 = 〈e1 + e2〉 C001 = 〈e1 − e2〉
C100 = 〈u1 + v1〉 C010 = 〈u2 + v2〉
C101 = 〈u1 − v1〉 C011 = 〈u2 − v2〉
81
C110 = 〈u3 + v3〉 C111 = 〈u3 − v3〉 (1)
This grading verifies that every homogeneous element is invertible.
Any other group grading on C is obtained by coarsening of these fine gradings.
3.2 Gradings on g2
Applying the previous procedure for translating gradings from C to g2 = L we can obtain the
fine gradings on this algebra. These are the following:
a) The Z2-toral grading given by
L0,0 = 〈Du1,v1,Du2,v2
,Du3,v3〉
L1,0 = 〈Dv1,u2〉 L0,1 = 〈Dv2,v3
〉 L1,1 = 〈Dv1,v3〉
L1,2 = 〈Du1,u2〉 L1,3 = 〈Du1,v3
〉 L2,3 = 〈Du2,v3〉
L−1,0 = 〈Du1,v2〉 L0,−1 = 〈Du2,u3
〉 L−1,−1 = 〈Du1,u3〉
L−1,−2 = 〈Dv1,v2〉 L−1,−3 = 〈Dv1,u3
〉 L−2,−3 = 〈Dv2,u3〉
This is of course the root decomposition relative to the Cartan subalgebra h = L0,0. Moreover,
if Φ is the root system relative to h, and we take ∆ = α1, α2 the roots related to Dv1,u2
and Dv2,v3respectively, it is clear that ∆ is a basis of Φ such that Ln1,n2
= Ln1α1+n2α2.
All the homogeneous elements, except the ones belonging to L0,0, are nilpotent.
b) The Z3
2-nontoral grading is given by
L0,0,0 = 0 L0,0,1 = h
L0,1,0 = 〈c + f,B + G〉 L0,1,1 = 〈−c + f,B − G〉
L1,1,0 = 〈A + D, b + g〉 L1,1,1 = 〈−A + D,−b + g〉
L1,0,0 = 〈a + d,C + F 〉 L1,0,1 = 〈−a + d,C − F 〉
if we denote by A := Dv1,u2, a := Dv2,v3
, c := Dv1,v3, b := Du1,u2
, G := Du1,v3, F := Du2,v3
,
D := Du1,v2, d := Du2,u3
, f := Du1,u3, g := Dv1,v2
, B := Dv1,u3and C := Dv2,u3
, that is, a
collection of root vectors corresponding to the picture
11
111
111
111
11
1
111
111
11
111
1
F
A
B
C
D
G
a
bc
d
g f
Notice that each of the nonzero homogeneous components is a Cartan subalgebra, that is,
every homogeneous element is semisimple.
82
4 Gradings on J and Der(J)
The Albert algebra is the exceptional Jordan algebra of dimension 27, that is,
J = H3(C) = x = (xij) ∈ M3(C) | xij = xji
with product x · y := 1
2(xy + yx), where juxtaposition stands for the usual matrix product.
Denote, if x ∈ C, by
x(1) =
0 0 0
0 0 x
0 x 0
x(2) =
0 0 x
0 0 0
x 0 0
x(3) =
0 x 0
x 0 0
0 0 0
and denote by E1, E2 and E3 the three orthogonal idempotents given by the elementary
matrices e11, e22 and e33 respectively. The multiplication table of the commutative algebra J
may be summarized in the following relations:
E2
i= Ei, Ei a(i) = 0, a(i)b(i) = 1
2tr(ab)(Ej + Ek),
EiEj = 0, Ei a(j) = 1
2a(j), a(i)b(j) = 1
2(ba)(k),
where (i, j, k) is any cyclic permutation of (1, 2, 3) and a, b ∈ C.
We fix for further reference our standard basis of the Albert algebra:
B = (E1, E2, E3, e(3)
1, e
(3)
2, u
(3)
1, u
(3)
2, u
(3)
3, v
(3)
1, v
(3)
2, v
(3)
3, e
(2)
2, e
(2)
1,−u
(2)
1,−u
(2)
2,
−u(2)
3,−v
(2)
1,−v
(2)
2,−v
(2)
3, e
(1)
1, e
(1)
2, u
(1)
1, u
(1)
2, u
(1)
3, v
(1)
1, v
(1)
2, v
(1)
3).
The Lie algebra f4 = Der(J) is generated by the set of derivations [Rx, Ry] | x, y ∈ C,
where Rx is the multiplication operator. As in the case of C and g2 we also have an algebraic
group isomorphism relating automorphisms of the Albert algebra J and of f4. This is given
by
Ad: aut(J) → aut(f4)
f 7→ Ad(f); Ad(f)(d) := fdf−1.
This provides also a mechanism to translate gradings from J to f4 and conversely. Since
Ad(f)([Rx, Ry]) = [Rf(x), Rf(y)], the grading induced on f4 by J = ⊕g∈GJg is given by
f4 = L = ⊕g∈GLg with Lg =∑
g1+g2=g[RJg1
, RJg2].
4.1 Gradings on the Albert algebra
There are four fine gradings on J , all of them quite natural if we look at them from a suitable
perspective.
a) The Z4-toral grading on J.
83
Define the maximal torus T0 of F4 whose elements are the automorphisms of J which are
diagonal relative to B. This is isomorphic to (F×)4 and it is not difficult to check that the
matrix of any such automorphism relative to B is
diag(
1, 1, 1, α,1
α, β, γ,
δ2
αβ γ,1
β,1
γ,α β γ
δ2, δ,
1
δ,α β
δ,α γ
δ,
δ
β γ,
δ
αβ,
δ
α γ,β γ
δ,δ
α,
α
δ,β
δ,γ
δ,
δ
α β γ,δ
β,δ
γ,α β γ
δ
)
for some α, β, γ, δ ∈ F×. Define now tα,β,γ,δ as the automorphism in T0 whose matrix relative
to B is just the above one. Notice that we have a Z4-grading on J such that tα,β,γ,δ acts in
J(n1,n2,n3,n4)with eigenvalue αn1βn2γn3δn4 . This is just
J0,0,0,0 = 〈E1, E2, E3〉
J1,0,0,0 = 〈e(3)
1〉 J0,0,0,−1 = 〈e
(2)
1〉 J−1,0,0,1 = 〈e
(1)
1〉
J−1,0,0,0 = 〈e(3)
2〉 J0,0,0,1 = 〈e
(2)
2〉 J1,0,0,−1 = 〈e
(1)
2〉
J0,1,0,0 = 〈u(3)
1〉 J1,1,0,−1 = 〈u
(2)
1〉 J0,1,0,−1 = 〈u
(1)
1〉
J0,0,1,0 = 〈u(3)
2〉 J1,0,1,−1 = 〈u
(2)
2〉 J0,0,1,−1 = 〈u
(1)
2〉
J−1,−1,−1,2 = 〈u(3)
3〉 J0,−1,−1,1 = 〈u
(2)
3〉 J−1,−1,−1,1 = 〈u
(1)
3〉
J0,−1,0,0 = 〈v(3)
1〉 J−1,−1,0,1 = 〈v
(2)
1〉, J0,−1,0,1 = 〈v
(1)
1〉
J0,0,−1,0 = 〈v(3)
2〉 J−1,0,−1,1 = 〈v
(2)
2〉 J0,0,−1,1 = 〈v
(1)
2〉
J1,1,1,−2 = 〈v(3)
3〉 J0,1,1,−1 = 〈v
(2)
3〉 J1,1,1,−1 = 〈v
(1)
3〉.
Recall from Schafer ([9, (4.41), p. 109]) that any x ∈ J satisfies a cubic equation x3 −
Tr(x)x2 + Q(x)x−N(x)1 = 0 where Tr(x),Q(x),N(x) ∈ F . Notice that again in this grading
every homogeneous element b /∈ J0,0,0,0 verifies that N(b) = 0.
b) The Z5
2-nontoral grading on J.
Define H ≡ H3(F ) = x ∈ M3(F ) | x = xt and K ≡ K3(F ) = x ∈ M3(F ) | x = −xt.
There is a vector space isomorphism
J = H ⊕ K ⊗ C0 (2)
given by Ei 7→ Ei, 1(i) 7→ 1(i) ∈ H and for x ∈ C0, x(i) 7→ (ejk − ekj) ⊗ x ∈ K ⊗ C0, being
(i, j, k) any cyclic permutation of (1, 2, 3) and eij ∈ M3(F ) the elementary (i, j)-matrix. Thus
J is a Jordan subalgebra of M3(F )⊗C with the product (c⊗x) · (d⊗ y) = 1
2((c⊗x)(d⊗ y)+
(d ⊗ y)(c ⊗ x)) for (c ⊗ x)(d ⊗ y) = cd ⊗ xy.
This way of looking at J allows us to observe that the gradings on the Cayley algebra C,
so as the gradings on the Jordan algebra H3(F ), induce gradings on J , because aut(C) and
aut(H3(F )) are subgroups of aut(J). Moreover, both kind of gradings are compatible. To be
more precise, consider a G1-grading on the Jordan algebra H = ⊕g∈G1Hg. This grading will
come from a grading on M3(F ) such that the Lie algebra K has also an induced grading ([7,
84
p. 184-185]). Taking a G2-grading on the Cayley algebra C = ⊕g∈G2Cg, we get a G1 × G2-
grading on J given by
Jg1,e = Hg1⊕ Kg1
⊗ (C0)e, Jg1,g2= Kg1
⊗ (C0)g2, (3)
for g1 ∈ G1, g2 ∈ G2 and e the zero element in any group.
Take now as G2-grading on g2 the Z3
2-nontoral grading (1). Note that (C0)e = 0, so that
Jg1,e = Hg1. And take as G1-grading on H3(F ), the Z
2
2-grading given by
He = 〈E1, E2, E3〉 H0,1 = 〈1(1)〉
H1,0 = 〈1(2)〉 H1,1 = 〈1(3)〉,
which induces in K the Z2
2-grading such that
Ke = 0 K0,1 = 〈e12 − e21〉 K1,1 = 〈e23 − e32〉 K1,0 = 〈e13 − e31〉.
Combining them as above, we find a Z5
2-grading on J with dimensions
dim Je,e = dim He = 3,
dim Je,g2= 0,
dim Jg1,e = dim Hg1= 1,
dim Jg1,g2= dim Kg1
⊗ (C0)g2= 1,
which is of type (24, 0, 1). The grading so obtained turns out to be, after using the isomor-
phism (2),
Je,e = 〈E1, E2, E3〉
Je,g = 0
J1,1,g = (Cg)(1)
J1,0,g = (Cg)(2)
J0,1,g = (Cg)(3),
for g ∈ Z3
2and Cg again given by (1).
c) The Z3
2× Z-nontoral grading on J.
If p ∈ SO(3, F ), denote by In(p) the automorphism in aut(H3(F )) given by In(p)(x) =
pxp−1. It is well known that a maximal torus of SO(3) is given by the matrices of the form
pα,β :=
1 0 0
0 α β
0 −β α
with α, β ∈ F such that α2 +β2 = 1. Thus, the set of all τα,β = In(pα,β) is a maximal torus of
aut(H3(F )) and the set of eigenvalues of τα,β is Sα,β = (α+iβ)n | n = 0,±1,±2. Supposing
85
|Sα,β| = 5, we find for τα,β the following eigenspaces
H2 = 〈−iE2 + iE3 + 1(1)〉
H1 = 〈−i(3) + 1(2)〉
H0 = 〈E1, E2 + E3〉
H−1 = 〈i(3) + 1(2)〉
H−2 = 〈iE2 − iE3 + 1(1)〉,
where the subindex n indicates that the eigenvalue of τα,β is (α+ iβ)n. This gives a Z-grading
on H3(F ), which induces the Cartan grading on K = Der(H3(F )), a Lie algebra of type a1.
An equivalent, but more comfortable, way of looking at these gradings, is:
H2 = 〈e23〉 K2 = 0
H1 = 〈e13 + e21〉 K1 = 〈e13 − e21〉
He = 〈E1, E2 + E3〉 Ke = 〈E2 − E3〉
H−1 = 〈e12 + e31〉 K−1 = 〈e12 − e31〉
H−2 = 〈e32〉 K−2 = 0
(4)
When mixing them with the Z3
2-grading on C as explained in (3), we obtain a Z×Z
3
2-fine
grading on J , whose dimensions are
dimJ2,e = dimH2 = 1
dimJ1,e = dimH1 = 1
dimJe,e = dim He = 2
dimJ2,g = 0
dimJ1,g = dim K1 ⊗ (C0)g = 1
dimJe,g = dim Ke ⊗ (C0)g = 1,
for g ∈ Z3
2. It is obviously a grading of type (25, 1). A detailed description of the components
could be obtained directly by using (3).
d) The Z3
3-nontoral grading on J.
There is another way in which the Albert algebra can be constructed, the so called Tits
construction described in [8, p. 525]. Let us start with the F -algebra A = M3(F ) and denote
by TrA,QA,NA : A → F the coefficients of the generic minimal polynomial such that x3 −
TrA(x)x2 + QA(x)x − NA(x)1 = 0 for all x ∈ A. Define also the quadratic map ♯ : A → A
by x♯ := x2 − TrA(x)x + QA(x)1. For any x, y ∈ A denote x × y := (x + y)♯ − x♯ − y♯, and
x∗ := 1
2x × 1 = 1
2TrA(x)1 − 1
2x. Finally consider the Jordan algebra A+ whose underlying
vector space agrees with that of A but whose product is x · y = 1
2(xy + yx). Next, define in
A3 := A × A × A the product
(a1, b1, c1)(a2, b2, c2) :=(a1 · a2 + (b1c2)
∗ + (b2c1)∗, a∗
1b2 + a∗
2b1 + 1
2(c1 × c2), c2a
∗1+ c1a
∗2+ 1
2(b1 × b2)
).
86
Then A3 with this product is isomorphic to J = H3(C).
This construction allows us to extend any f automorphism of A to the automorphism
f• of J = A3 given by f•(x, y, z) := (f(x), f(y), f(z)). As a further consequence we will
be able to get gradings on J coming from gradings on the associative algebra A via this
monomorphism of algebraic groups. Thus, consider the Z2
3-grading on A produced by the
commuting automorphisms f := In(p) and g := In(q), for p = diag(1, ω, ω2), being ω a
primitive cubic root of the unit and
q =
(0 1 0
0 0 1
1 0 0
)
.
The simultaneous diagonalization of A relative to f, g yields A = ⊕2
i,j=0Ai,j where
A00 = 〈1A〉, A01 = 〈ω2e11 − ωe22 + e33〉, A02 = 〈−ωe11 + ω2e22 + e33〉,
A10 = 〈e13 + e21 + e32〉, A11 = 〈ω2e13 − ωe21 + e32〉, A12 = 〈−ωe13 + ω2e21 + e32〉,
A20 = 〈e12 + e23 + e31〉, A21 = 〈ω2e12 − ωe23 + e31〉, A22 = 〈−ωe12 + ω2e23 + e31〉.
If we make a simultaneous diagonalization of J relative to the automorphisms f•, g• we
get the toral Z2
3-grading J = ⊕2
i,j=0A3
i,j. Now consider a third order three automorphism
φ ∈ aut(J) given by φ(a0, a1, a2) = (a0, ωa1, ω2a2). It is clear that f•, g•, φ is a com-
mutative set of semisimple automorphisms of J , producing the simultaneous diagonalization
J = ⊕2
i,j,k=0Ji,j,k where
Ji,j,0 = Aij × 0 × 0
Ji,j,1 = 0 × Aij × 0
Ji,j,2 = 0 × 0 × Aij ,
so that we have 27 one-dimensional homogeneous components. In particular this Z3
3-grading
on J is fine and nontoral (otherwise J0,0,0 would contain three orthogonal idempotents).
Observe also that the homogeneous elements are invertible.
4.2 Gradings on f4
Recall that the isomorphism Ad: aut(J) ∼= aut(f4), introduced at the beginning of Section 4,
provides a mechanism for translating gradings from J to f4 and conversely. Therefore, there
will be four fine gradings on f4 too, over Z4, Z
5
2, Z
3
2× Z and Z
3
3.
a) The Z4-toral grading on f4.
Denote by ωi the i-th element in the basis B. The Z4-toral grading on J = ⊕g∈Z4Jg
87
induces a grading on f4 = L = ⊕g∈Z4Lg by Lg =∑
g1+g2=g[RJg1
, RJg2]:
L0,0,0,0 = 〈[Rω4 , Rω5 ], [Rω6 , Rω9 ], [Rω7 , Rω10 ], [Rω12 , Rω13 ]〉
L1,0,0,0 = 〈[Rω1 , Rω4 ]〉 L0,0,0,1 = 〈[Rω1 , Rω13 ]〉 L0,1,1,0 = 〈[Rω6 , Rω7 ]〉
L−1,−2,−1,2 = 〈[Rω8 , Rω9 ]〉 L1,0,0,−1 = 〈[Rω2 , Rω21 ]〉 L0,1,1,−1 = 〈[Rω1 , Rω19 ]〉
L−1,−1,0,2 = 〈[Rω7 , Rω8 ]〉 L−1,−1,0,1 = 〈[Rω1 , Rω17 ]〉 L1,1,1,−1 = 〈[Rω2 , Rω27 ]〉
L0,1,1,−2 = 〈[Rω5 , Rω11 ]〉 L0,−1,0,1 = 〈[Rω2 , Rω25 ]〉 L1,1,1,−2 = 〈[Rω1 , Rω11 ]〉
L−1,−1,0,0 = 〈[Rω5 , Rω9 ]〉 L0,−1,0,0 = 〈[Rω1 , Rω9 ]〉 L2,1,1,−2 = 〈[Rω4 , Rω11 ]〉
L−1,0,1,0 = 〈[Rω5 , Rω7 ]〉 L0,0,1,0 = 〈[Rω1 , Rω7 ]〉 L1,−1,0,0 = 〈[Rω4 , Rω9 ]〉
L0,0,1,−1 = 〈[Rω2 , Rω23 ]〉 L1,0,1,0 = 〈[Rω4 , Rω7 ]〉 L1,0,1,−1 = 〈[Rω1 , Rω15 ]〉
L1,0,1,−2 = 〈[Rω9 , Rω11 ]〉 L1,1,2,−2 = 〈[Rω7 , Rω11 ]〉 L0,−1,1,0 = 〈[Rω7 , Rω9 ]〉
L−1,0,0,0 = 〈[Rω1 , Rω5 ]〉 L0,0,0,−1 = 〈[Rω1 , Rω12 ]〉 L0,−1,−1,0 = 〈[Rω9 , Rω10 ]〉
L1,2,1,−2 = 〈[Rω6 , Rω11 ]〉 L−1,0,0,1 = 〈[Rω2 , Rω20 ]〉 L0,−1,−1,1 = 〈[Rω1 , Rω16 ]〉
L1,1,0,−2 = 〈[Rω10 , Rω11 ]〉 L1,1,0,−1 = 〈[Rω1 , Rω14 ]〉 L−1,−1,−1,1 = 〈[Rω2 , Rω24 ]〉
L0,−1,−1,2 = 〈[Rω4 , Rω8 ]〉 L0,1,0,−1 = 〈[Rω2 , Rω22 ]〉 L−1,−1,−1,2 = 〈[Rω1 , Rω8 ]〉
L1,1,0,0 = 〈[Rω4 , Rω6 ]〉 L0,1,0,0 = 〈[Rω1 , Rω6 ]〉 L−2,−1,−1,2 = 〈[Rω5 , Rω8 ]〉
L1,0,−1,0 = 〈[Rω4 , Rω10 ]〉 L0,0,−1,0 = 〈[Rω1 , Rω10 ]〉 L−1,1,0,0 = 〈[Rω5 , Rω6 ]〉
L0,0,1,−1 = 〈[Rω2 , Rω26 ]〉 L−1,0,−1,0 = 〈[Rω5 , Rω10 ]〉 L−1,0,−1,1 = 〈[Rω1 , Rω18 ]〉
L−1,0,−1,2 = 〈[Rω6 , Rω8 ]〉 L−1,−1,−2,2 = 〈[Rω8 , Rω10 ]〉 L0,1,−1,0 = 〈[Rω6 , Rω10 ]〉.
This is of course the root decomposition relative to the Cartan subalgebra h = L0,0,0,0. If Φ
is the root system relative to h, and we take ∆ = α1, α2, α3, α4 the set of roots related to
[Rω8, Rω9
], [Rω6, Rω7
], [Rω1, Rω13
] and [Rω1, Rω4
] respectively, it is straightforward to check
that ∆ is a basis of Φ such that the root space Ln1α1+n2α2+n3α3+n4α4and the homogeneous
component Ln4−n1,n2−n1,n2−n1,n3+2n1coincide.
b) The Z5
2-nontoral grading on f4.
We got the Z5
2-grading on J by looking at J as H ⊕ K ⊗ C0. But f4 is its algebra of
derivations, hence there should exist some model of f4 in terms of H, K and C. In fact we
can see f4 as
L = Der(C) ⊕ K ⊕ H0 ⊗ C0
identifying Der(H3(F )) in a natural way with K in the known Tits unified construction for
the Lie exceptional algebras (for instance, see [9, p. 122]).
Consider a G1-grading on the Jordan algebra H = ⊕g∈G1Hg. This grading will come from
a grading on M3(F ) so that the Lie algebra K has also an induced grading. Take now the Z3
2-
grading on the Cayley algebra C = ⊕g∈G2=Z3
2
Cg and the induced grading Der(C) = ⊕g∈G2Ng.
All this material induces a G1 × G2-grading on L by means of
Lg1,e = Kg1, Le,g2
= Ng2⊕ (H0)e ⊗ (C0)g2
, Lg1,g2= (H0)g1
⊗ (C0)g2, (5)
which is just the grading induced by the G1 × G2-grading on J described by (3).
In the case of the Z5
2-grading recall that G1 = Z
2
2,
He = 〈E1, E2, E3〉 H0,1 = 〈e12 + e21〉 H1,1 = 〈e23 + e32〉 H1,0 = 〈e13 + e31〉
Ke = 0 K0,1 = 〈e12 − e21〉 K1,1 = 〈e23 − e32〉 K1,0 = 〈e13 − e31〉
88
and dim(C0)g = 1, dimNg = 2 for all g ∈ Z3
2\ (0, 0, 0). Therefore
dim Le,e = 0,
dim Le,g2= dim Ng2
+ dim(H0)e ⊗ (C0)g2= 4,
dim Lg1,e = dim Kg1= 1,
dim Lg1,g2= dim(H0)g1
⊗ (C0)g2= 1,
and so the grading is of type (24, 0, 0, 7), with all the homogeneous elements semisimple.
c) The Z3
2× Z-nontoral grading on f4.
We obtain the grading by the method just explained, with the G1 = Z-grading on H and
K described in (4), and by crossing it with the Z3
2-grading on C. In such a way we get a
Z3
2× Z-grading of type (31, 0, 7), since
dimL2,e = 0,
dimL1,e = dim K1 = 1,
dimLe,e = dimKe = 1,
dimL2,g = dim H2 ⊗ (C0)g = 1,
dimL1,g = dim H1 ⊗ (C0)g = 1,
dimLe,g = dim Ng + dim(H0)e ⊗ (C0)g = 3,
and Lg is dual to L−g, so they have the same dimensions.
The detailed description of the components of the last two gradings can be made by using
(5), but it is not worth to be developed here.
d) The Z3
3-nontoral grading on f4.
The easiest way to visualize this grading intrinsically, that is, with no reference to a
particular basis or computer methods, is probably looking at the automorphisms inducing
the grading. Adams gave a construction of the Lie algebra e6 from three copies of a2 ([1,
p. 85]). Once the automorphisms have been given in e6 we will restrict them to f4.
Given a 3-dimensional F -vector space X in which a nonzero alternate trilinear map
det: X × X × X → F has been fixed, we can identify the exterior product with the dual
space by X∧X≈→ X∗ such that x∧y 7→ det(x, y,−) ∈ hom(X,F ). And in a dual way we can
identify X∗ ∧ X∗ with X through det∗, the dual map of det. Consider three 3-dimensional
vector spaces Xi (i = 1, 2, 3), and define:
L = sl(X1) ⊕ sl(X2) ⊕ sl(X3) ⊕ X1 ⊗ X2 ⊗ X3 ⊕ X∗1 ⊗ X∗
2 ⊗ X∗3 ,
endowed with a Lie algebra structure with the product
[⊗fi,⊗xi] =∑
k=1,2,3
i6=j 6=k
fi(xi)fj(xj)(fk(−)xk − 1
3fk(xk)idXk
)
[⊗xi,⊗yi] = ⊗(xi ∧ yi)
[⊗fi,⊗gi] = ⊗(fi ∧ gi)
89
for any xi, yi ∈ Xi, fi, gi ∈ X∗i, with the wedge products as above, and where the actions of
the Lie subalgebra∑
sl(Xi) on X1⊗X2⊗X3 and X∗1⊗X∗
2⊗X∗
3are the natural ones (the i-th
simple ideal acts on the i-th slot). The Lie algebra L is isomorphic to e6. The decomposition
L = L0 ⊕ L1 ⊕ L2 is a Z3-grading for L0 = sl(X1) ⊕ sl(X2) ⊕ sl(X3), L1 = X1 ⊗ X2 ⊗ X3
and L2 = X∗1⊗ X∗
2⊗ X∗
3. Take φ1 the automorphism which induces the grading, that is,
φ1|Li= ωiidLi
for ω a primitive cubic root of the unit. We are giving two automorphisms
commuting with φ1.
A family of automorphisms of the Lie algebra commuting with φ1 is the following. If
ρi : Xi → Xi, i = 1, 2, 3, are linear maps preserving det: X3
i→ F , the linear map ρ1 ⊗ ρ2 ⊗
ρ3 : L1 → L1 can be uniquely extended to an automorphism of L such that its restriction to
sl(Vi) ⊂ L0 is the conjugation map g 7→ ρigρ−1
i.
Fix now basis u0, u1, u2 of X1, v0, v1, v2 of X2, and w0, w1, w2 of X3 with det(u0, u1, u2) =
det(v0, v1, v2) = det(w0, w1, w2) = 1. Consider φ2 the unique automorphism of e6 extending
the map
ui ⊗ vj ⊗ wk 7→ ui+1 ⊗ vj+1 ⊗ wk+1
(indices module 3). Finally let φ3 be the unique automorphism of e6 extending the map
ui ⊗ vj ⊗ wk 7→ ωiui ⊗ ωjvj ⊗ ωkwk = ωi+j+kui ⊗ vj ⊗ wk.
The set φi3
i=1is a commutative set of semisimple automorphisms, and it induces a Z
3
3-
grading on e6. The grading is nontoral since its zero homogeneous component is null.
Some computations prove that the rest of the homogeneous components are all of them
3-dimensional.
The nice 3-symmetry described in e6 is inherited by f4. Indeed graphically speaking,
f4 arises by folding e6. More precisely, taking X2 = X3 we can consider on e6 the unique
automorphism τ : e6 → e6 extension of u ⊗ v ⊗ w 7→ u ⊗ w ⊗ v. This is an order two
automorphism commuting with the previous φi for i = 1, 2, 3. The subalgebra of elements
fixed by τ is
sl(X1) ⊕ sl(X2) ⊕ X1 ⊗ Sym2(X2) ⊕ X∗1 ⊗ Sym2(X∗
2 ),
where SymnXi denotes the symmetric powers. This is a simple Lie algebra of dimension 52,
hence f4. Furthermore, denoting also by φi : f4 → f4 the restriction of the corresponding auto-
morphisms of e6, the set φi3
i=1is a set of commuting semisimple order three automorphisms
of f4 with no fixed points other than 0. So it induces a nontoral Z3
3-grading on f4 of type
(0, 26), with all the homogeneous elements semisimple.
References
[1] J. F. Adams. Lectures on Exceptional Lie Groups. The University Chicago Press, 1984.
90
[2] Y. A. Bahturin, I. P. Shestakov and M. V. Zaicev. Gradings on simple Jordan and Lie
algebras. Journal of Algebra 283 (2005), 849-868.
[3] C. Draper and C. Martın. Gradings on g2. Linear Algebra and its Applications 418, no 1,
(2006), 85-111.
[4] C. Draper and C. Martın. Gradings on the Albert algebra and on f4. Preprint Jordan
Theory Preprint Archives 232, March 2007.
[5] M. Havlcek, J. Patera and E. Pelatonova. On Lie gradings II. Linear Algebra and its
Applications 277 (1998), 97-125.
[6] A. Elduque. Gradings on Octonions. Journal of Algebra 207 (1998), 342-354.
[7] N. Jacobson. Structure and Representations of Jordan algebras. Amer. Math. Soc. Colloq.
Publ., vol 39, Amer. Math. Soc., Providence, RI, 1969.
[8] M-A. Knus, A. Merkurjev, M. Rost, J-P. Tignol. The Book of involutions. American
Mathematical Society Colloquium Publications, Vol 44, 1998.
[9] R. D. Schafer. An introduction to nonassociative algebras. Academic Press. New York,
San Francisco, London, 1966.
91
92
Notes on a general compactification of symmetric
spaces
Vadim Kaimanovich∗ Pedro J. Freitas†
Abstract
In this paper we outline the results regarding a construction of a general K-equivariant
compactification of the symmetric space G/K, from a compactification of the Weyl cham-
ber. A more detailed paper on this subject is expected soon. 1
Symmetric spaces are a classical object of the Riemannian geometry and serve as a testing
ground for numerous concepts and notions. The simplest symmetric space is the hyperbolic
plane which can be naturally compactfied by the circle at infinity, which is essentailly the
only reasonable compactifictaion of the hyperbolic plane. However, for higher rank sym-
metric spaces the situation is more complicated, and there one can define several different
comapctifications—the visibility, the Furstenberg, the Martin, the Karpelevich—to name just
the most popular ones. The present report is a part of an ongoing project aimed at under-
standing the nature and structure of general compactifications of symmetric spaces.
1 General Concepts
We start by defining the notation (either well known or taken from [GJT], with minor ad-
justments) and the concepts necessary. The results that follow can be found in [GJT] and
[He].
We take G, a semisimple connected Lie group with finite center, and let K be a maximal
compact subgroup. Denote by g and k the Lie algebras of G and K respectively.
Let g = k ⊕ p be the Cartan decomposition of g, p being the orthogonal complement of
k in g, with respect to the Killing form B. The space p can be identified with the tangent
space to G/K at the coset K, which we’ll denote by o. The restriction of the Killing form to
this space is positive definite, and thus provides an inner product in p.
∗IRMAR, Universite Rennes-1, Campus de Beaulieu, 35042 Rennes Cedex, France†Center for Linear and Combinatorial Structures, University of Lisbon, Av. Prof. Gama Pinto 2, 1649-003
Lisbon, Portugal12000 Mathematics Subject Classification. Primary: 53C35, 54D35.
Key words and phrases. Compactifications, Symmetric Spaces.
93
We take a to be a fixed Cartan subalgebra of p, a+ a fixed Weyl chamber, Σ the set of all
the roots of g with respect to a (the so-called restricted roots), Σ+ the set of positive roots,
∆ the set of the simple roots. We denote by d be the rank of G (the dimension of a).
The action of G on G/K is by left multiplication. Every element of G can be written as
k. exp(X) with k ∈ K and X ∈ p—this is an easy consequence of the Cartan decomposition,
which states that every element of G can be expressed as k1 exp(X)k2, k1, k2 ∈ K, X ∈ a+
— as the closed Weyl chamber will be a very important object, we’ll denote w := a+. From
this we can easily conclude that every point in G/K can be presented as k exp(X).o, k ∈ K,
X ∈ w, which means that the K-orbit of exp(w), is whole symmetric space. Moreover, the
element X ∈ w is uniquely defined, and is called the generalized radius. The element k is
unique modulo the stabilizer of X for the adjoint action of K over w.
Given a topological group H, we’ll say that a topological space A is an H-space if there
is an action of H on A (which we’ll denote by a dot) and the map
H × A → A
(h, a) 7→ h.a
is continuous. If B is another H-space, and φ : A → B is a continuous map, we say that φ is
H-equivariant if, for any h ∈ H,a ∈ A, φ(h.a) = h.φ(a).
If B is compact, φ is an embedding, and φ(A) is dense in B, we’ll say that (φ,B) (or
simply B if there is no confusion about the map involved) is a compactification of A. If φ is
H-equivariant, we’ll say that B is an H-compactification.
2 Building the compactification
We are now concerned with the definition of a compactification of the space G/K via com-
pactifications of the closed Weyl chamber. There are a few compactifications of G/K that
can be presented this way, as we will see later.
Now suppose we have a compactification of w, w, that is Hausdoff and satisfies the fol-
lowing condition:
(⋆) For sequences xn, x′n∈ w, If xn → x ∈ ∂w and d(xn, x′
n) → 0, then x′
n→ x.
Notice that if x ∈ w, we always have this property, since the topology in w is given by
the metric d.
Now, we are looking for a K-invariant compactification of G/K that restricted to w will
be w. We present a process of doing this.
94
Consider the space K × w, and the map π1 : K × w → G/K defined naturally by
π1(k, x) := k. exp(x).o. Consider also the compact space K × w and its quotient by the
equivalence relation ∼ defined by the following rules:
(i) for x, y ∈ w, (k, x) ∼ (r, y) ⇔ x = y and r exp(x).o = s exp(y).o;
(ii) for x, y ∈ ∂w, (k, x) ∼ (r, y) ⇔ there exist convergent sequences (k, xn) and (r, yn) in
K × w with lim xn = x, lim yn = y, such that d(k exp(xn), r exp(yn)) → 0.
The relation ∼ is clearly an equivalence relation. Denote by K the quotient space.
Notation. Condition (i) assures that there is a bijection between G/K and (K ×w)/ ∼,
name it ι. We will therefore use the notation kE(x) to denote (k, x)/ ∼, for k ∈ K and x ∈ w.
Thus, ι(k exp(x).o) = kE(x) in K.
Now take the inclusion and projection maps
ι1 : K × w → K × w π2 : K × w → K.
We have that following diagram commutes.
K × wπ1−→ G/K
yι1
yι
K × wπ2−→ K
It can be proved that the image of K ×w is dense in K, and that ι is an embedding. This
makes K into a compactification of G/K, since K is clearly compact. The following result is
a parallel to the polar decomposition on G/K.
Proposition 2.1. For x, y ∈ w, and k, r ∈ K, we have that kE(x) = rE(y) if and only if
x = y and k−1r ∈ StabK(E(x)). In particular, the “generalized radius” x is well defined,
even when kE(x) ∈ ∂K.
This furthers the analogy with elements in G/K. The compactification has the following
properties.
Proposition 2.2. 1. The space K is Hausdorff.
2. The projection π2 : K × w → K is a closed map.
3. The space K is a K-space, and ι is K-equivariant.
Moreover, one can prove, using certain classes of converging sequences (just as it is done
in [GJT]) that this compactification has some uniqueness properties.
Theorem 2.3. The compactification K has the following properties.
95
1. It is a K-compactification.
2. When restricted to w is w.
3. It respects intersections of Weyl chambers.
4. It is metrizable if w is metrizable.
5. Being metrizable, it dominates any other compactification satisfying conditions 1–3.
Examples. There are a few known compactifications that are particular cases of our
compactification K, originating from different compactifications of w. Among these are the
compactifications of Furstenberg, Martin, Satake and Karpelevich.
3 An action of G and independence of the base point.
In the case G = SL(n, R) and K = SO(n), it is possible to define an action of G on K. So, in
this section, we take G = SL(n, R), plus the following assumptions on the compactification
w.
1. We’ll assume that, in the compactification of the Cartan subalgebra a, if a sequence an
converges, then for any a ∈ a, a + an also converges.
2. The compactification w is a refinement of the Furstenberg compactification.
For g ∈ G, denote by Kg the compactification obtained by apllying the above process to
G/K, but using g.o as a reference point, instead of o.
Under these conditions, it is possible to identify a point of K with a point of Kg, comparing
the behaviour of sequences in both compactifications, and the rules of convergence for both—
and prove they are the same. The proof available so far is quite technical, and the authors
are working on a better version. We thus can prove the following result.
Proposition 3.1. With the notation above, K and Kg are the same compactification.
This allows us to define an action of G on K, as follows: given a point x ∈ G/K, and
g ∈ G, we take g.x to be the point in K corresponding to the point g.x ∈ Kg.
References
[Ba] W. Ballmann, Lectures on Spaces of Nonpositive Curvature, Birkhauser Verlag, 1995.
[FH] W. Fulton and J. Harris, Representation Theory: a First Course, Springer-Verlag,
New York, 1991.
96
[GJT] Y. Guivarc’h, L. Ji, J. C. Taylor, Compactifications of Symmetric Spaces, Birkhauser,
1998.
[He] S. Helgason, Differential Geometry, Lie Groups and Symmetric Spaces, Academic
Press, 1978.
[Ke] J. Kelley, General Topology, D. Van Nostrand Co. Inc., 1957.
[Li] E. Lima, Topologia dos Espacos Metricos, Notas de Matematica do IMPA, 1958.
97
98
Tridiagonal pairs, the q-tetrahedron algebra, Uq(sl2),
and Uq(sl2)
Darren Funk-Neubauer∗
Abstract
Let V denote a finite dimensional vector space over an algebraically closed field. A
tridiagonal pair is an ordered pair A, A∗ of diagonalizable linear transformatons on V such
that (i) the eigenspaces of A can be ordered as Vidi=0 with A∗Vi ⊆ Vi−1+Vi+Vi+1 for 0 ≤
i ≤ d; (ii) the eigenspaces of A∗ can be ordered as V ∗
i di=0 with AV ∗
i ⊆ V ∗
i−1 + V ∗
i + V ∗
i+1
for 0 ≤ i ≤ d; (iii) there are no nonzero proper subspaces of V which are invariant
under both A and A∗. Tridiagonal pairs arise in the representation theory of various Lie
algebras, associative algebras, and quantum groups. We recall the definition of one such
algebra called the q-tetrahedron algebra and discuss its relation to the quantum groups
Uq(sl2) and Uq(sl2). We discuss the role the q-tetrahedron algebra plays in the attempt to
classify tridiagonal pairs. In particular, we state a theorem which connects the actions of
a certain type of tridiagonal pair A, A∗ on V to an irreducible action of the q-tetrahedron
algebra on V .
1 Tridiagonal Pairs
In this paper we discuss the connection between tridiagonal pairs and representation theory.
However, tridiagonal pairs originally arose in algebraic combinatorics through the study of
a combinatorial object called a P- and Q-polynomial association scheme [4]. In addition,
tridiagonal pairs are related to many other areas of mathematics. For example, they appear
in the study of orthogonal polynomials and special functions [12], the theory of partially
ordered sets [11], and statistical mechanics [13]. We now define a tridiagonal pair.
Definition 1.1. [4] Let V denote a finite dimensional vector space over an algebraically closed
field K. A tridiagonal pair on V is an ordered pair A,A∗ where A : V → V and A∗ : V → V
are linear transformatons that satisfy the following conditions:
(i) Each of A,A∗ is diagonalizable.
∗Department of Mathematics, University of Wisconsin-Madison, Van Vleck Hall, 480 Lincoln Drive, Madi-
son, WI 53706-1388, USA. E-mail: [email protected]
99
(ii) The eigenspaces of A can be ordered as Vid
i=0with A∗Vi ⊆ Vi−1+Vi+Vi+1 (0 ≤ i ≤ d),
where V−1 = 0, Vd+1 = 0.
(iii) The eigenspaces of A∗ can be ordered as V ∗iδ
i=0with AV ∗
i⊆ V ∗
i−1+ V ∗
i+ V ∗
i+1(0 ≤
i ≤ δ), where V ∗−1
= 0, V ∗δ+1
= 0.
(iv) There does not exist a subspace W ⊆ V such that AW ⊆ W , A∗W ⊆ W , W 6= 0,
W 6= V .
According to a common notational convention A∗ denotes the conjugate-transpose of A. We
am not using this convention; the linear transformations A,A∗ are arbitrary subject to (i)–
(iv) above.
Referring to Definition 1.1, it turns out d = δ [4]; we call this common value the diameter of
the tridiagonal pair. We call an ordering of the eigenspaces of A (resp. A∗) standard when-
ever it satisfies (ii) (resp. (iii)) above. We call an ordering of the eigenvalues of A (resp. A∗)
standard whenever the corresponding ordering of the eigenspaces of A (resp. A∗) is standard.
The tridiagonal pairs for which the Vi, V∗i
all have dimension 1 are called Leonard pairs. The
Leonard pairs are classified and correspond to a family of orthogonal polynomials consisting of
the q-Racah polynomials and related polynomials in the Askey scheme [12]. Currently there
is no classification of tridiagonal pairs. We will discuss the connection between tridiagonal
pairs, the q-tetrahedron algebra, and the quantum groups Uq(sl2) and Uq(sl2). The hope is
that this connection will eventually lead to a classification of tridiagonal pairs.
2 The eigenvalues of a tridiagonal pair
In this section we describe how the standard orderings of the eigenvalues of a tridiagonal pair
satisfy a certain three term recurrance relation.
Let A,A∗ denote a tridiagonal pair on V . Let θid
i=0denote a standard ordering of the
eigenvalues of A. Let θ∗id
i=0denote a standard ordering of the eigenvalues of A∗.
Theorem 2.1. [4, Theorem 11.1] The expressions
θi−2 − θi+1
θi−1 − θi
,θ∗i−2
− θ∗i+1
θ∗i−1
− θ∗i
are equal and independent of i for 2 ≤ i ≤ d − 1.
We now describe the solutions to the recurrance from Theorem 2.1.
100
Theorem 2.2. [4, Theorem 11.2] Solving the recurrence in Theorem 2.1 we have the following.
For some scalars q, a, b, c, a∗, b∗, c∗ ∈ K the sequences θid
i=0and θ∗
id
i=0have one of the
following forms:
Case I: For 0 ≤ i ≤ d
θi = a + bqi + cq−i,
θ∗i
= a∗ + b∗qi + c∗q−i.
Case II: For 0 ≤ i ≤ d
θi = a + bi + ci(i − 1)/2,
θ∗i
= a∗ + b∗i + c∗i(i − 1)/2.
Case III: The characteristic of K is not 2, and for 0 ≤ i ≤ d
θi = a + b(−1)i + ci(−1)i,
θ∗i
= a∗ + b∗(−1)i + c∗i(−1)i.
For the remainder of this paper we will be concerned with the tridiagonal pairs whose eigen-
values are as in Case I from Theorem 2.2. Such tridiagonal pairs are closely connected to
representations of the quantum groups Uq(sl2) and Uq(sl2). The study of this connection
inspired the definition of the q-tetrahedron algebra.
3 The q-tetrahedron algebra, Uq(sl2), and Uq(sl2)
In this section we define the q-tetrahedron algebra and discuss its connection to the quantum
groups Uq(sl2) and Uq(sl2).
For the remainder of the paper we will assume q ∈ K is nonzero and not a root of unity.
We will use the following notation. For an integer i ≥ 0 we define
[i] =qi − q−i
q − q−1and [i]! = [i][i − 1] · · · [2][1].
We interpret [0]! = 1.
Let Z4 = Z/4Z denote the cyclic group of order 4.
Definition 3.1. [7, Definition 6.1] Let ⊠q denote the unital associative K-algebra that has
generators
xij | i, j ∈ Z4, j − i = 1 or j − i = 2
and the following relations:
101
(i) For i, j ∈ Z4 such that j − i = 2,
xijxji = 1.
(ii) For h, i, j ∈ Z4 such that the pair (i − h, j − i) is one of (1, 1), (1, 2), (2, 1),
qxhixij − q−1xijxhi
q − q−1= 1.
(iii) For h, i, j, k ∈ Z4 such that i − h = j − i = k − j = 1,
x3
hixjk − [3]qx
2
hixjkxhi + [3]qxhixjkx
2
hi− xjkx
3
hi= 0. (1)
We call ⊠q the q-tetrahedron algebra.
We now recall the definition of Uq(sl2).
Definition 3.2. [9, p. 9] Let Uq(sl2) denote the unital associative K-algebra with generators
K±1, e± and the following relations:
KK−1 = K−1K = 1,
Ke±K−1 = q±2e±,
e+e− − e−e+ =K − K−1
q − q−1.
We now recall an alternate presentation for Uq(sl2).
Lemma 3.3. [8, Theorem 2.1] The algebra Uq(sl2) is isomorphic to the unital associative
K-algebra with generators x±1, y, z and the following relations:
xx−1 = x−1x = 1,
qxy − q−1yx
q − q−1= 1,
qyz − q−1zy
q − q−1= 1,
qzx − q−1xz
q − q−1= 1.
We now present a lemma which relates Uq(sl2) and ⊠q.
Lemma 3.4. [7, Proposition 7.4] For i ∈ Z4 there exists a K-algebra homomorphism from
Uq(sl2) to ⊠q that sends
x → xi,i+2, x−1 → xi+2,i, y → xi+2,i+3, z → xi+3,i.
We now recall the definition of Uq(sl2).
102
Definition 3.5. [1, Definition 2.2] Let Uq(sl2) denote the unital associative K-algebra gen-
erated by K±1
i, e±
i, i ∈ 0, 1 subject to the relations
KiK−1
i= K−1
iKi = 1, (2)
K0K1 = K1K0, (3)
Kie±iK−1
i= q±2e±
i, (4)
Kie±jK−1
i= q∓2e±
j, i 6= j, (5)
e+
ie−i− e−
ie+
i=
Ki − K−1
i
q − q−1, (6)
e±0e∓1
= e∓1e±0, (7)
(e±i)3e±
j− [3](e±
i)2e±
je±i
+ [3]e±ie±j(e±
i)2 − e±
j(e±
i)3 = 0, i 6= j. (8)
We now recall an alternate presentation for Uq(sl2).
Theorem 3.6. [5, Theorem 2.1], [10] The algebra Uq(sl2) is isomorphic to the unital asso-
ciative K-algebra with generators xi, yi, zi, i ∈ 0, 1 and the following relations:
x0x1 = x1x0 = 1,
qxiyi − q−1yixi
q − q−1= 1,
qyizi − q−1ziyi
q − q−1= 1,
qzixi − q−1xizi
q − q−1= 1,
qziyj − q−1yjzi
q − q−1= 1, i 6= j,
y3
iyj − [3]qy
2
iyjyi + [3]qyiyjy
2
i− yjy
3
i= 0, i 6= j,
z3
izj − [3]qz
2
izjzi + [3]qzizjz
2
i− zjz
3
i= 0, i 6= j.
We now present a lemma which relates Uq(sl2) and ⊠q.
Lemma 3.7. [7, Proposition 8.3] For i ∈ Z4 there exists a K-algebra homomorphism from
Uq(sl2) to ⊠q that sends
x1 → xi,i+2, y1 → xi+2,i+3, z1 → xi+3,i,
x0 → xi+2,i, y0 → xi,i+1, z0 → xi+1,i+2.
103
4 The connection between tridiagonal pairs and ⊠q
We now recall a theorem which connects ⊠q to the tridiagonal pairs whose eigenvalues are as
in Case I of Theorem 2.2 (with a = a∗ = c = b∗ = 0 and b, c∗ 6= 0).
Theorem 4.1. [6, Theorem 2.7], [7, Theorem 10.4] Let A,A∗ denote a tridiagonal pair on
V . Let θid
i=0(resp. θ∗
id
i=0) denote a standard ordering of the eigenvalues of A (resp.
A∗). Assume there exist nonzero scalars b, c∗ ∈ K such that θi = bq2i−d and θ∗i
= c∗qd−2i for
0 ≤ i ≤ d. Then there exists a unique irreducible representation of ⊠q on V such that bx0
acts as A and c∗x2 acts as A∗.
Since the finite dimensional irreducible representations of ⊠q are completely understood [7]
Theorem 4.1 classifies tridiagonal pairs where the eigenvalues of A (resp. A∗) are θi = bq2i−d
(resp. θ∗i
= c∗qd−2i).
Given Theorem 4.1 it is natural to ask the following question. If the eigenvalues of A and
A∗ are more general can we still construct a representation of ⊠q on V ? More specifically,
if the eigenvalues of A are θi = bq2i−d for 0 ≤ i ≤ d and the eigenvalues of A∗ are θ∗i
=
b∗q2i−d + c∗qd−2i for 0 ≤ i ≤ d can we construct a representation of ⊠q on V that generalizes
the construction in Theorem 4.1? We answer this question in the next section.
5 A generalization of Theorem 4.1
In this section we present a theorem which connects ⊠q to the tridiagonal pairs whose eigen-
values are as in Case I of Theorem 2.2 (with a = a∗ = c = 0 and b, b∗, c∗ 6= 0).
Before we state this theorem we have a number of prelimanary definitions.
Let A,A∗ denote a tridiagonal pair on V and let Vid
i=0(resp. V ∗
id
i=0) denote a standard
ordering of the eigenspace of A (resp. A∗). For 0 ≤ i ≤ d define Ui = (V ∗0
+· · ·+V ∗i)∩(Vi+· · ·+
Vd). It turns out each of Uid
i=0is nonzero and V is their direct sum [4]. The sequence Ui
d
i=0
is called the split decomposition of A,A∗. There exist linear transformations R : V → V and
L : V → V such that (i) U0, . . . , Ud are the common eigenspaces for A − R,A∗ − L and (ii)
RUi ⊆ Ui+1 and LUi ⊆ Ui−1 for 0 ≤ i ≤ d [4]. R (resp. L) is called the raising (resp.
lowering) map associated to A,A∗.
Definition 5.1. [3] Let V denote a finite dimensional vector space over K. Let A,A∗ denote
a tridiagonal pair on V and let θid
i=0(resp. θ∗
id
i=0) denote a standard ordering of the
eigenvalues of A (resp. A∗). Assume for nonzero b ∈ K that θi = bq2i−d for 0 ≤ i ≤ d.
Furthermore, assume for nonzero b∗, c∗ ∈ K that θ∗i
= b∗q2i−d + c∗qd−2i for 0 ≤ i ≤ d. Let
Uid
i=0denote the split decomposition of A,A∗. Let R (resp. L) denote the raising (resp.
104
lowering) map associated to A,A∗. It is known that dim(U0) = 1. Thus for 0 ≤ i ≤ d the
space U0 is an eigenspace of LiRi; let σi denote the coresponding eigenvalue.
The following polynomial will be used to state our theorem. It is a slight modification of the
Drinfeld polynomial which is well known in representation theory [1, 2].
Definition 5.2. [3] With reference to Definition 5.1 define the polynomial P ∈ K[λ] by
P (λ) =
d∑
i=0
qi(1−i) σi λi
[i]! 2,
The following theorem uses P to explain the connection between finite dimensional irreducible
representations of ⊠q and the tridiagonal pairs whose eigenvalues are as in Case I of Theorem
2.2 (with a = a∗ = c = 0 and b, b∗, c∗ 6= 0).
Theorem 5.3. [3] With reference to Definition 5.1 and Definition 5.2, the following are
equivalent:
(i) There exists a representation of ⊠q on V such that bx01 acts as A and b∗x30 + c∗x23
acts as A∗.
(ii) P (q2d−2(q − q−1)−2) 6= 0.
Suppose (i),(ii) hold. Then the ⊠q representation on V is unique and irreducible.
References
[1] V. Chari, A. Pressley, Quantum affine algebras, Commun. Math. Phys. 142 (1991) 261–
283.
[2] V.G. Drinfeld, A new realization of Yangians and quantized affine algebras, Soviet Math.
Dokl. 36 (1988) 212–216.
[3] D. Funk-Neubauer, Tridiagonal pairs and the q-tetrahedron algebra, in preparation.
[4] T. Ito, K. Tanabe, P. Terwilliger, Some algebra related to P- and Q-polynomial association
schemes, in: Codes and Association Schemes (Piscataway NJ, 1999), Amer. Math. Soc.,
Providence, RI, 2001, pp. 167–192.
[5] T. Ito, P. Terwilliger, Tridiagonal pairs and the quantum affine algebra Uq(sl2), Ramanu-
jan J., to appear. arXiv:math.QA0310042.
[6] T. Ito, P. Terwilliger, Two non-nilpotent linear transformations that satisfy the cubic
q-Serre relations, J. Algebra Appl., submitted. arXiv:math.QA050839.
105
[7] T. Ito, P. Terwilliger, The q-tetrahedron algebra and its finite dimensional modules,
Comm. Algebra, submitted. arXiv:math.QA0602199.
[8] T. Ito, P. Terwilliger, and C. W. Weng, The quantum algebra Uq(sl2) and its equitable
presentation, J. Algebra, submitted.
[9] J.C. Jantzen, Lectures on quantum groups, Amer. Math. Soc., Providence RI, 1996.
[10] P. Terwilliger, The equitable presentation for the quantum group Uq(g) associated with
a symmetrizable Kac-Moody algebra g, J. Algebra, submitted.
[11] P. Terwilliger, The incidence algebra of a uniform poset, Math and its applications, 20
(1990) 193–212.
[12] P. Terwilliger, Two linear transformations each tridiagonal with respect to an eigenbasis
of the other, Linear Algebra Appl., 330 (2001) 149–203.
[13] P. Terwilliger, Two relations that generalize the q-Serre relations and the Dolan-Grady
relations. In Physics and Combinatorics. Proceedings of the Nagoya 1999 International
Workshop (Nagoya, Japen 1999), World Scientific Publishing Co., Inc., River Edge, NJ,
2001.
106
Quantum Lie algebras via modified reflection
equation algebra
Dimitri Gurevich∗ Pavel Saponov†
1 Introduction
A Lie super-algebra was historically the first generalization of the notion of a Lie algebra. Lie
super-algebras were introduced by physicists in studying dynamical models with fermions.
In contrast with the usual Lie algebras defined via the classical flip P interchanging any two
elements P (X ⊗ Y ) = Y ⊗ X, the definition of a Lie super-algebra is essentially based on a
super-analog of the permutation P . This super-analog is defined on a Z2-graded vector space
V = V0⊕ V
1where 0, 1 ∈ Z2 is a ”parity”. On homogeneous elements (i.e. those belonging
to either V0
or V1) its action is P (X ⊗ Y ) = (−1)XY Y ⊗X, where X stands for the parity of
a homogeneous element X ∈ V .
Then a Lie super-algebra is the following data
(g = g0⊕ g
1, P : g ⊗ g → g ⊗ g, [ , ] : g ⊗ g → g) ,
where g is a super-space, P is a super-flip, and [ , ] is a Lie super-bracket, i.e. a linear operator
which is subject to three axioms:
1. [X,Y ] = −(−1)XY [Y,X];
2. [X, [Y,Z]] + (−1)X(Y +Z)[Y, [Z,X]] + (−1)Z(X+Y )[Z, [X,Y ]] = 0;
3. [X,Y ] = X + Y .
Here X,Y,Z are assumed to be arbitrary homogenous elements of g. Note that all axioms
can be rewritten via the corresponding super-flip. For instance the axiom 3 takes the form
P (X ⊗ [Y,Z]) = [ , ]12P23P12(X ⊗ Y ⊗ Z).
(As usual, the indices indicate the space(s) where a given operator is applied.)
In this paper we discuss the problem what is a possible generalization of the notion of a
Lie super-algebra related to ”flips” of more general type.
∗USTV, Universite de Valenciennes, 59304 Valenciennes, France. E-mail: [email protected]†Division of Theoretical Physics, IHEP, 142284 Protvino, Russia. E-mail: [email protected]
107
The first generalization of the notion of a Lie super-algebra was related to gradings dif-
ferent from Z2. The corresponding Lie type algebras were called Γ-graded ones (cf. [Sh]).
The next step was done in [G1] where there was introduced a new generalization of the
Lie algebra notion related to an involutive symmetry defined as follows. Let V be a vector
space over a ground field K (usually C or R) and R : V ⊗2 → V ⊗2 be a linear operator. It is
called a braiding if it satisfies the quantum Yang-Baxter equation
R12R23R12 = R23R12R23 ,
where R12 = R ⊗ I, R23 = I ⊗ R are operators in the space V ⊗3. If such a braiding satisfies
the condition R2 = I (resp., (R−q I)(R+q−1 I) = 0, q ∈ K) we call it an involutive symmetry
(resp., a Hecke symmetry). In the latter case q is assumed to be generic1.
Two basic examples of generalized Lie algebras are analogs of the Lie algebras gl(n) and
sl(n) (or of their super-analogs gl(m|n) and sl(m|n)). They can be associated to any ”skew-
invertible” (see Section 2) involutive symmetry R : V ⊗2 → V ⊗2. We denote them gl(VR)
and sl(VR) respectively. The generalized Lie algebras gl(VR) and sl(VR) are defined in the
space End (V ) of endomorphisms of the space V . Their enveloping algebras U(gl(VR)) and
U(sl(VR)) (which can be defined in a natural way) are equipped with a braided Hopf structure
such that the coproduct coming in its definition acts on the generators X ∈ gl(VR) or sl(VR)
in the classical manner: ∆ : X → X ⊗ 1 + 1 ⊗ X.
Moreover, if an involutive symmetry R is a deformation of the usual flip (or super-flip) the
enveloping algebras U(gl(VR)) and U(sl(VR)) are deformations of their classical (or super-)
counterparts.
There are known numerous attempts to define a quantum (braided) Lie algebra similar
to generalized ones but without assuming R to be involutive. Let us mention some of them:
[W], [LS], [DGG], [GM]. In this note we compare the objects defined there with gl type Lie
algebras-like objects introduced recently in [GPS]. Note that the latter objects can be associ-
ated with any skew-invertible Hecke symmetry, in particular, that related to Quantum Groups
(QG) of An series. Their enveloping algebras are treated in terms of the modified reflection
equation algebra (mREA) defined bellow. These enveloping algebras have good deformation
properties and the categories of their finite dimensional equivariant representations look like
those of the Lie algebras gl(m|n). Moreover, these algebras can be equipped with a structure
of braided bi-algebras. Though the corresponding coproduct acts on the generators of the
algebras in a non-classical way it is in a sense intrinsic (it has nothing in common with the
coproduct in the QGs). Moreover, it allows to define braided analogs of (co)adjoint vectors
fields.
1Note that there exists a big family of Hecke and involutive symmetries which are not deformations of the
usual flip (cf. [G2]). Even the Poincare-Hilbert (PH) series corresponding to the ”symmetric” Sym(V ) =
T (V )/〈Im (qI −R)〉 and ”skew-symmetric”∧
(V ) = T (V )/〈Im (q−1I + R)〉 algebras can drastically differ from
the classical ones, whereas the PH series are stable under a deformation.
108
We think that apart from generalized Lie algebras related to involutive symmetries (de-
scribed in Section 2) there is no general definition of a quantum (braided) Lie algebra. More-
over, reasonable quantum Lie algebras exist only for the An series (or more generally, for
any skew-invertible Hecke symmetry). As for the quantum Lie algebras of the Bn, Cn, Dn
series introduced in [DGG], their enveloping algebras are not deformations of their classical
counterparts and for this reason they are somewhat pointless objects.
2 Generalized Lie algebras
Let R : V ⊗2 → V ⊗2 be an involutive symmetry. Then the data
(V,R, [ , ] : V ⊗2 → V )
is called a generalized Lie algebra if the following holds
1. [ , ]R(X ⊗ Y ) = −[X,Y ];
2. [ , ] [ , ]12(I + R12R23 + R23R12)(X ⊗ Y ⊗ Z) = 0;
3. R[ , ]12(X ⊗ Y ⊗ Z) = [ , ]12R23R12(X ⊗ Y ⊗ Z).
Such a generalized Lie algebra is denoted g.
Note that the generalized Jacobi identity (the axiom 2) can be rewritten in one of the
following equivalent forms
[ , ] [ , ]23(I + R12R23 + R23R12)(X ⊗ Y ⊗ Z) = 0;
[ , ] [ , ]12(X(Y ⊗ Z − R(Y ⊗ Z))) = [X, [Y,Z]];
[ , ] [ , ]23((X ⊗ Y − R(X ⊗ Y ))Z) = [[X,Y ], Z].
Example 1. If R is the the usual flip then the third axiom is fulfilled automatically and we
get a usual Lie algebra. If R is a super-flip then we get a Lie super-algebra. In the both cases
R is involutive.
The enveloping algebras of the generalized Lie algebra g can be defined in a natural way:
U(g) = T (V )/〈X ⊗ Y − R(X ⊗ Y ) − [X,Y ]〉 .
(Hereafter 〈I〉 stands for the ideal generated by a set I.) Let us introduce the symmetric
algebra Sym (g) of the generalized Lie algebra g by the same formula but with 0 instead of
the bracket in the denominator of the above formula.
For this algebra there exists a version of the Poincare-Birhoff-Witt theorem.
Theorem 2. The algebra U(g) is canonically isomorphic to Sym(g).
109
A proof can be obtained via the Koszul property established in [G2] and the results of
[PP]. Also, note that similarly to the classical case this isomorphism can be realized via a
symmetric (w.r.t. the symmetry R) basis.
Definition 3. We say that a given braiding R : V ⊗2 → V ⊗2 is skew-invertible if there exists
a morphism Ψ : V ⊗2 → V ⊗2 such that
Tr2Ψ12R23 = P13 = Tr2Ψ23R12
where P is the usual flip.
If R is a skew-invertible braiding, a ”categorical significance” can be given to the dual
space of V . Let V ∗ be the vector space dual to V . This means that there exist a non-
degenerated pairing 〈 , 〉 : V ∗ ⊗ V → K and an extension of the symmetry R to the space
(V ∗⊕V )⊗2 → (V ∗⊕V )⊗2 (we keep the same notation for the extended braiding) such that the
above pairing is R-invariant. This means that on the space V ∗⊗ V ⊗W (resp., W ⊗V ∗ ⊗V )
where either W = V or W = V ∗ the following relations hold
R 〈 , 〉12 = 〈 , 〉23 R12 R23 (resp., R 〈 , 〉23 = 〈 , 〉12 R23 R12) .
(Here as usual, we identify X ∈ W with X ⊗ 1 and 1 ⊗ X.)
Note that if such an extension exists it is unique. By fixing bases xi ∈ V and xi⊗xj ∈ V ⊗2
we can identify the operators R and Ψ with matrices ‖Rkl
ij‖ and ‖Ψkl
ij‖ respectively. For
example,
R(xi ⊗ xj) = Rkl
ijxk ⊗ xl
(from now on we assume the summation over the repeated indices).
Then the above definition can be presented in the following matrix form
Rkl
ijΨ jn
lm= δk
mδn
i.
If ix is the left dual basis of the space V ∗, i.e. such that 〈 jx, xi〉 = δj
ithen we put
〈xi,jx〉 = 〈 , 〉Ψjl
ik
kx ⊗ xl = Cj
i, where C
j
i= Ψjk
ik.
(Note that the operator Ψ is a part of the braiding R extended to the space (V ∗⊕V )⊗2.) By
doing so, we ensure R-invariance of the pairing V ⊗ V ∗ → K.
As shown in [GPS] for any skew-invertible Hecke symmetry R the following holds
Cj
iBk
j= q−2aδk
i, where B
j
i= Ψkj
ki
with an integer a depending on the the HP series of the algebra Sym(V ) (see footnote 1).
So, if q 6= 0 the operators C and B (represented by the matrices ‖Cj
i‖ and ‖Bk
j‖ respectively)
are invertible. Therefore, we get a non-trivial pairing
〈 , 〉 : (V ⊕ V ∗)⊗2 → K
110
which is R-invariant.
Note that these operators B and C can be introduced without fixing any basis in the
space V as follows
B2 = Tr(1)(Ψ12), C1 = Tr(2)(Ψ12). (1)
Let us exhibit an evident but very important property of these operators
Tr(1)(B1 R12) = I, T r(2)(C2 R12) = I. (2)
By fixing the basis hj
i= xi ⊗
jx in the space End (V ) ∼= V ⊗ V ∗ equipped with the usual
product
: End (V )⊗2 → End (V )
we get the following multiplication table hj
i hl
k= δ
j
khl
i.
Below we use another basis in this algebra, namely that lj
i= xi ⊗ xj where xj is the right
dual basis in the space V ∗, i.e. such that 〈xi, xj〉 = δ
j
i. Note that the multiplication table for
the the product in this basis is lj
i lm
k= B
j
klmi
(also see formula (6)).
Let R be the above extension of a skew-invertible braiding to the space (V ∗⊕V )⊗2. Then
a braiding REnd (V ) : End (V )⊗2 → End (V )⊗2 can be defined in a natural way:
REnd (V ) = R23 R34R12 R23 ,
where we used the isomorphism End (V ) ∼= V ⊗ V ∗.
Observe that the product in the space End (V ) is R-invariant and therefore REnd (V )-
invariant. Namely, we have
REnd (V )(X Y,Z) = 23(REnd (V ))12(REnd (V ))23(X ⊗ Y ⊗ Z) ,
REnd (V )(X,Y Z) = 12(REnd (V ))23(REnd (V ))12(X ⊗ Y ⊗ Z) .
Example 4. Let R : V ⊗2 → V ⊗2 be a skew-invertible involutive symmetry. Define a gener-
alized Lie bracket by the rule
[X,Y ] = X Y − REnd (V )(X ⊗ Y ) .
Then the data (End (V ), REnd (V ), [ , ]) is a generalized Lie algebra (denoted gl(VR)).
Besides, define the R-trace TrR : End (V ) → K as follows
TrR(hj
i) = Bi
jh
j
i, X ∈ End (V ) .
The R-trace possesses the following properties : The pairing
End (V ) ⊗ End (V ) → K : X ⊗ Y 7→ 〈X,Y 〉 = TrR(X Y )
is non-degenerated;
111
It is REnd (V )-invariant in the following sense
REnd (V )((TrRX) ⊗ Y ) = (I ⊗ TrR)REnd (V )(X ⊗ Y ) ,
REnd (V )(X ⊗ (TrRY )) = (TrR ⊗ I)REnd (V )(X ⊗ Y ) ; TrR [ , ] = 0.
Therefore the set X ∈ gl(VR)|TrR X = 0 is closed w.r.t. the above bracket. Moreover,
this subspace squared is invariant w.r.t the symmetry REnd (V ). Therefore this subspace
(denoted sl(VR)) is a generalized Lie subalgebra.
Observe that the enveloping algebra of any generalized Lie algebra possesses a braided
Hopf algebra structure such that the coproduct ∆ and antipode S are defined on the generators
in the classical way
∆(X) = X ⊗ 1 + 1 ⊗ X, S(X) = −X .
For details the reader is referred to [G2].
Also, observe that while R is a super-flip the generalized Lie algebra gl(VR) (resp., sl(VR))
is nothing but the Lie super-algebras gl(m|n) (resp., sl(m|n)).
3 Quantum Lie algebras for Bn, Cn, Dn series
In this Section we restrict ourselves to the braidings coming from the QG Uq(g) where g is a
Lie algebra of one of the series Bn, Cn, Dn. By the Jacobi identity, the usual Lie bracket
[ , ] : g ⊗ g → g
is a g-morphism.
Let us equip the space g with a Uq(g) action which is a deformation of the usual adjoint
one. The space g equipped with such an action is denoted gq. Our immediate goal is to define
an operator
[ , ]q : gq ⊗ gq → gq
which would be a Uq(g)-covariant deformation of the initial Lie bracket. This means that the
q-bracket satisfies the relation
[ , ]q(a1(X) ⊗ a2(Y )) = a([X ⊗ Y ]q) ,
where a is an arbitrary element of the QG Uq(g), a1 ⊗ a2 = ∆(a) is the Sweedler notation for
the QG coproduct ∆, and a(X) stands for the result of applying the element a ∈ Uq(g) to an
element X ∈ gq.
112
Let us show that Uq(g)-covariance of the bracket entails its R-invariance where R =
P πg⊗g(R) is the image of the universal quantum R-matrix R composed with the flip P .
Indeed, due to the relation
∆12(R) = R13R23 ,
we have (by omitting πg⊗g)
R[ , ]1(X ⊗ Y ⊗ Z) = PR([X,Y ] ⊗ Z) = P [ , ]12∆12R(X ⊗ Y ⊗ Z) =
P [ , ]12R13R23(X ⊗ Y ⊗ Z) = P [ , ]12P13R13P23R23(X ⊗ Y ⊗ Z) =
P [ , ]12P13P23R12R23(X ⊗ Y ⊗ Z) = [ , ]23R12R23(X ⊗ Y ⊗ Z) .
Finally, we have
R[ , ]12 = [ , ]23R12R23, R[ , ]23 = [ , ]12R23R12
(the second relation can be obtained in a similar way).
Thus, the Uq(g)-covariance of the bracket [ , ]q can be considered as an analog of the axiom
3 from the above list. In fact, if g belongs to one of the series Bn, Cn or Dn, this property
suffices for unique (up to a factor) definition of the bracket [ , ]q. Indeed, in this case it is
known that if one extends the adjoint action of g to the space g⊗ g (via the coproduct in the
enveloping algebra), then the latter space is multiplicity free with respect to this action. This
means that there is no isomorphic irreducible g-modules in the space g⊗ g. In particular, the
component isomorphic to g itself appears only in the skew-symmetric subspace of g ⊗ g. A
similar property is valid for decomposition of the space gq⊗gq into a direct sum of irreducible
Uq(g)-modules (recall that q is assumed to be generic).
Thus, the map [ , ]q, being a Uq(g)-morphism, must kill all components in the decom-
position of gq ⊗ gq into a direct sum of irreducible gq-submodules except for the component
isomorphic to gq. Being restricted to this component, the map [ , ]q is an isomorphism. This
property uniquely defines the map [ , ]q (up to a non-zero factor). For an explicit computation
of the structure constants of the q-bracket [ , ]q the reader is referred to the paper [DGG].
Note that the authors of that paper embedded the space gq in the QG Uq(g). Nevertheless,
it is possible to do all the calculations without such an embedding but using the QG just as
a substitute of the corresponding symmetry group.
Now, we want to define the enveloping algebra of a quantum Lie algebra gq. Since the
space gq ⊗ gq is multiplicity free, we conclude that there exists a unique Uq(g)-morphism
Pq : gq ⊗ gq → gq ⊗ gq which is a deformation of the usual flip and such that P 2
q= I. Indeed,
in order to introduce such an operator it suffices to define q-analogs of symmetric and skew-
symmetric components in gq ⊗ gq. Each of them can be defined as a direct sum of irreducible
Uq(g)-submodules of gq ⊗gq which are q-counterparts of the Uq(g)-modules entering the usual
symmetric and skew-symmetric subspaces respectively.
113
Now, the enveloping algebra can be defined as a quotient
U(gq) = T (gq)/〈X ⊗ Y − Pq(X ⊗ Y ) − [ , ]q〉 .
Thus, we have defined the quantum Lie algebra gq and its enveloping algebra related
to the QG of Bn, Cn, Dn series. However, the question what properties of these quantum
Lie algebras are similar to those of generalized Lie algebras is somewhat pointless since the
algebra U(gq) is not a deformation of its classical counterpart. Moreover, its ”q-commutative”
analog (which is defined similarly to the above quotient but without the q-bracket [ , ]q in the
denominator) is not a deformation of the algebra Sym(g). For the proof, it suffices to verify
that the corresponding semiclassical term is not a Poisson bracket. (However, it becomes
Poisson bracket upon restriction to the corresponding algebraic group.)
Remark 5. A similar construction of a quantum Lie algebra is valid for any skew-inver-
tible braiding of the Birman-Murakami-Wenzl type. But for the same reason it is out of our
interest.
Also, note that the Lie algebra sl(2) possesses a property similar to that above: the
space sl(2) ⊗ sl(2) being equipped with the extended adjoint action is a multiplicity free
sl(2)-module. So, the corresponding quantum Lie algebra and its enveloping algebra can be
constructed via the same scheme. However, the latter algebra is a deformation of its classical
counterpart. This case is consider in the next Sections as a part of our general construction
related to Hecke symmetries.
4 Modified reflection equation algebra and its representation
theory
In this section we shortly describe the modified reflection equation algebra (mREA) and the
quasitensor Schur-Weyl category of its finite dimensional equivariant representations. Our
presentation is based on the work [GPS], where these objects were considered in full detail.
The starting point of all constructions is a Hecke symmetry R. As was mentioned in
Introduction, the Hecke symmetry is a linear operator R : V ⊗2 → V ⊗2, satisfying the quantum
Yang-Baxter equation and the additional Hecke condition
(R − q I)(R + q−1I) = 0 ,
where a nonzero q ∈ K is generic, in particular, is not a primitive root of unity. Besides, we
assume R to be skew-invertible (see Definition 3).
Fixing bases xi ∈ V and xi ⊗ xj ∈ V ⊗2, 1 ≤ i, j ≤ N = dimV , we identify R with a
N2 × N2 matrix ‖Rkl
ij‖. Namely, we have
R(xi ⊗ xj) = Rkl
ijxk ⊗ xl , (3)
114
where the lower indices label the rows of the matrix, the upper ones — the columns.
As is known, the Hecke symmetry R allows to define a representations ρR of the Ak−1
series Hecke algebras Hk(q), k ≥ 2, in tensor powers V ⊗k:
ρR : Hk(q) → End(V ⊗k) ρR(σi) = Ri := I⊗(i−1) ⊗ R ⊗ I⊗(k−i−1) ,
where elements σi, 1 ≤ i ≤ k − 1 form the set of the standard generators of Hk(q).
The Hecke algebra Hk(q) possesses the primitive idempotents eλ
a∈ Hk(q), which are in
one-to-one correspondence with the set of all standard Young tableaux (λ, a), corresponding
to all possible partitions λ ⊢ k. The index a labels the tableaux of a given partition λ in
accordance with some ordering.
Under the representation ρR, the primitive idempotents eλ
aare mapped into the projection
operators
Eλ
a(R) = ρR(eλ
a) ∈ End (V ⊗k) , (4)
these projectors being some polynomials in Ri, 1 ≤ i ≤ k − 1.
Under the action of these projectors the spaces V ⊗k, k ≥ 2, are expanded into the direct
sum
V ⊗k =⊕
λ⊢k
dλ⊕
a=1
V(λ,a), V(λ,a) = Im(Eλ
a) , (5)
where the number dλ stands for the total number of the standard Young tableaux, which can
be constructed for a given partition λ.
Since the projectors Eλ
awith different a are connected by invertible transformations, all
spaces V(λ,a) with fixed λ and different a are isomorphic. Note, that the isomorphic spaces
V(λ,a) (at a fixed λ) in decomposition (5) are treated as particular embeddings of the space
Vλ into the tensor product V ⊗k. Hereafter we use the notation Vλ for the class of the spaces
V(λ,a) equipped with one or another embedding in V ⊗k.
In a similar way we define classes V ∗µ. First, note that the Hecke symmetry being extended
to the space (V ∗)⊗2 is given in the basis xi ⊗ xj as follows
R(xi ⊗ xj) = Rji
lkxk ⊗ xl
(and similarly in the basis ix⊗ jx). It is not difficult to see that the operator R so defined in
the space (V ∗)⊗2 is a Hecke symmetry. Thus, by using the above method we can introduce
spaces V ∗(µ,a)
looking like those from (5) and define the classes V ∗µ.
Now, let us define a rigid quasitensor Schur-Weyl category SW(V ) whose objects are spaces
Vλ and V ∗µ
labelled by partitions of nonnegative integers, as well as their tensor products
Vλ ⊗ V ∗µ
and all finite sums of these spaces.
Among the morphisms of the category SW(V ) are the above left and right pairings and
the set of braidings RU,W : U⊗W → W ⊗U for any pair of objects U and W . These braidings
can be defined in a natural way. In order to define them on a couple of objects of the form
115
Vλ ⊗ V ∗µ
we embed them into appropriate products V ⊗k ⊗ (V ∗)⊗l and define the braiding
RU,W as an appropriate restriction. Note, that all these braidings are R-invariant maps (cf.
[GPS] for detail). Note that the category SW(V ) is monoidal quasitensor rigid according to
the standard terminology (cf. [CP]).
Now we are aiming at introducing modified reflection equation algebra and equipping
objects of the category SW(V ) with a structure of its modules.
Again, consider the space End (V ) equipped with the basis lj
i(see Section 2). Note that
the element lj
iacts on the elements of the space V as follows
lj
i(xk) := xi 〈x
j , xk〉 = xiBj
k. (6)
Introduce the N × N matrix L = ‖l j
i‖. Also, define its ”copies” by the iterative rule
L1
:= L1 := L ⊗ I, Lk+1
:= RkLkR−1
k. (7)
Observe that the isolated spaces Lk
have no meaning (except for that L1). They can be only
correctly understood in the products L1L
2, L
1L
2L
3and so on, but this notation is useful in
what follows.
Definition 6. The associative algebra generated by the unit element eL and the indetermi-
nates lj
i1 ≤ i, j ≤ N subject to the following matrix relation
R12L1R12L1 − L1R12L1R12 − ~ (R12 L1 − L1 R12) = 0 , (8)
is called the modified reflection equation algebra (mREA) and denoted L(Rq, ~).
Note, that at ~ = 0 the above algebra is known as the reflection equation algebra L(Rq).
Actually, at q 6= ±1 one has L(Rq, ~) ∼= L(Rq). Since at ~ 6= 0 it is always possible to
renormalize generators L 7→ ~ L. So, below we consider the case ~ = 1.
Thus, the mREA is the quotient algebra of the free tensor algebra T (End (V )) over the
two-sided ideal, generated by the matrix elements of the left hand side of (8). It can be
shown, that the relations (8) are R-invariant, that is the above two-sided ideal is invariant
when commuting with any object U under the action of the braidings RU,End (V ) or REnd (V ),U
of the category SW(V).
Taking into account (2) one can easily prove, that the action (6) gives a basic (vector)
representation of the mREA L(Rq, 1) in the space V
ρ1(lj
i) ⊲ xk = xiB
j
k, (9)
where the symbol ⊲ stands for the (left) action of a linear operator onto an element. Since B
is non-degenerated, the representation is irreducible.
Another basic (covector) representation ρ∗1
: L(Rq, 1) → End (V ∗) is given by
ρ∗1(lj
i) ⊲ xk = −xr R
kj
ri. (10)
116
one can prove, that the maps End (V ) → End (V ) and End (V ) → End (V ∗) generated by
lj
i7→ ρ1(l
j
i) and l
j
i7→ ρ∗
1(l j
i)
are the morphisms of the category SW(V ).
Definition 7. A representation ρ : L(Rq, 1) → End (U) where U is an object of the category
SW(V ) is called equivariant if its restriction to End (V ) is a categorical morphism.
Thus, the above representations ρ1 and ρ∗1
are equivariant.
Note that there are known representations of the mREA which are nor equivariant. How-
ever, the class of equivariant representations of the mREA is very important. In particular,
because the tensor product of two equivariant L(Rq, 1)-modules can be also equipped with a
structure of an equivariant L(Rq, 1)-module via a ”braided bialgebra structure” of the mREA.
Let us briefly describe this structure. It consists of two maps: the braided coproduct ∆
and counit ε.
The coproduct ∆ is an algebra homomorphism of L(Rq, 1) into the associative algebra
L(Rq) which is defined as follows. As a vector space over the field K the algebra L(Rq) is isomorphic to the tensor product
of two copies of mREA
L(Rq) = L(Rq, 1) ⊗L(Rq, 1) . The product ⋆ : (L(Rq))⊗2 → L(Rq) is defined by the rule
(a1 ⊗ b1) ⋆ (a2 ⊗ b2) := a1a′2⊗ b′
1b2 , ai ⊗ bi ∈ L(Rq) , (11)
where a1a′2
and b1b′2
are the usual product of mREA elements, while a′1
and b′1
result
from the action of the braiding REnd(V ) (see Section 2) on the tensor product b1 ⊗ a2
a′2 ⊗ b′1 := REnd(V )(b1 ⊗ a2) . (12)
The braided coproduct ∆ is now defined as a linear map ∆ : L(Rq, 1) → L(Rq) with the
following properties:
∆(eL) := eL ⊗ eL
∆(lji) := l
j
i⊗ eL + eL ⊗ l
j
i− (q − q−1)
∑
klki⊗ l
j
k
∆(ab) := ∆(a) ⋆ ∆(b) ∀ a, b ∈ L(Rq, 1) .
(13)
In addition to (13), we introduce a linear map ε : L(Rq, 1) → K
ε(eL) := 1
ε(lji) := 0
ε(ab) := ε(a)ε(b) ∀ a, b ∈ L(Rq, 1) .
(14)
117
One can show (cf. [GPS]) that the maps ∆ and ε are indeed algebra homomorphisms and
that they satisfy the relation
(id ⊗ ε)∆ = id = (ε ⊗ id)∆ .
Let now U and W be two equivariant mREA-modules with representations ρU : L(Rq, 1) →
End (U) and ρW : L(Rq, 1) → End (W ) respectively. Consider the map ρU⊗W : L(Rq) →
End (U ⊗ W ) defined as follows
ρU⊗W (a ⊗ b) ⊲ (u ⊗ w) = (ρU (a) ⊲ u′) ⊗ (ρW (b′) ⊲ w) , a ⊗ b ∈ L(Rq) , (15)
where
u′ ⊗ b′ := REnd (V ),U (b ⊗ u) .
Definition (15) is self-consistent since the map b 7→ ρW (b′) is also a representation of the
mREA L(Rq, 1).
The following proposition holds true.
Proposition 8. ([GPS]) The action (15) defines a representation of the algebra L(Rq).
Note again, that the equivariance of the representations in question plays a decisive role
in the proof of the above proposition.
As an immediate corollary of the proposition 8 we get the rule of tensor multiplication of
equivariant L(Rq, 1)-modules.
Corollary 9. Let U and W be two L(Rq, 1)-modules with equivariant representations ρU and
ρW . Then the map L(Rq, 1) → End (U ⊗ W ) given by the rule
a 7→ ρU⊗W (∆(a)) , ∀ a ∈ L(Rq, 1) (16)
is an equivariant representation. Here the coproduct ∆ and the map ρU⊗W are given respec-
tively by formulae (13) and (15).
Thus, by using (16) we can extend the basic representations ρ1 and ρ∗1
to the represen-
tations ρk and ρ∗l
in tensor products V ⊗k and (V ∗⊗l) respectively. These representations are
reducible, and their restrictions on the representations ρλ,a in the invariant subspaces V(λ,a)
(see (5)) are given by the projections
ρλ,a = Eλ
a ρk (17)
and similarly for the subspaces V ∗(µ,a)
. By using (16) once more we can equip each object of
the category SW(V ) with the structure of an equivariant L(Rq, 1)-module.
118
5 Quantum Lie algebras related to Hecke symmetries
In this section we consider the question to which extent one can use the scheme of section 2 in
the case of non-involutive Hecke symmetry R for definition of the corresponding Lie algebra-
like object. For such an object related to a Hecke symmetry R we use the term quantum or
braided Lie algebra. Besides, we require the mREA, connected with the same symmetry R,
to be an analog of the enveloping algebra of the quantum Lie algebra. Finally, we compare
the properties of the above generalized Lie algebras and quantum ones.
Let us recall the interrelation of a usual Lie algebra g and its universal enveloping algebra
U(g). As is known, the universal enveloping algebra for a Lie algebra g is a unital associative
algebra U(g) possessing the following properties: There exists a linear map τ : g → U(g) such that 1 and Im τ generate the whole U(g). The Lie bracket [x, y] of any two elements of g has the image
τ([x, y]) = τ(x)τ(y) − τ(y)τ(x).
Let us rewrite these formulae in an equivalent form. Note that the tensor square g ⊗ g
splits into the direct sum of symmetric and skew symmetric components
g ⊗ g = gs ⊕ ga, gs = ImS, ga = ImA ,
where S and A are the standard (skew)symmetrizing operators
S(x ⊗ y) = x ⊗ y + y ⊗ x, A(x ⊗ y) = x ⊗ y − y ⊗ x ,
where we neglect the usual normalizing factor 1/2. Then the skew-symmetry property of the
classical Lie bracket is equivalent to the requirement
[ , ]S(x ⊗ y) = 0 . (18)
The image of the bracket in U(g) is presented as follows
τ([x, y]) = A (τ(x) ⊗ τ(y)) , (19)
where stands for the product in the associative algebra U(g). The Jacobi identity for the Lie bracket [ , ] translates into the requirement that the
correspondence x 7→ [x, ] generate the (adjoint) representation of U(g) in the linear
subspace τ(g) ⊂ U(g).
So, we define a braided Lie algebra as a linear subspace L1 = End (V ) of the mREA
L(Rq, 1), which generates the whole algebra and is equipped with the quantum Lie bracket.
We want the bracket to satisfy some skew-symmetry condition, generalizing (18), and define
119
a representation of the mREA in the same linear subspace L1 via an analog of the Jacobi
identity.
As L1, let us take the linear span of mREA generators
L1 = End (V ) ∼= V ⊗ V ∗ .
Together with the unit element this subspace generate the whole L(Rq, 1) by definition.
In order to find the quantum Lie bracket, consider a particular representation of L(Rq, 1)
in the space End (V ). In this case the general formula (16) reads
lj
i7→ ρV ⊗V ∗(∆(l j
i)) ,
where we should take the basic representations (9) and (10) as ρV (lji) and ρV ∗(lj
i) respectively.
Omitting straightforward calculations, we write the final result in the compact matrix form
ρV ⊗V ∗(L1) ⊲ L
2= L1R12 − R12L1 , (20)
where the matrix Lk
is defined in (7).
Let us define
[L1, L
2] = L1R12 − R12L1 . (21)
The generalized skew-symmetry (the axiom 1 from Section 2) of this bracket is now modified
as follows. In the space L1 ⊗L1 one can construct two projection operators Sq and Aq which
are interpreted as q-symmetrizer and q-skew-symmetrizer respectively (cf. [GPS]). Then
straightforward calculations show that the above bracket satisfies the relation
[ , ]Sq(L1⊗ L
2) = 0 , (22)
which is the generalized skew-symmetry condition, analogous to (18).
Moreover, if we rewrite the defining commutation relations of the mREA (8) in the equiv-
alent form
L1L
2− R−1
12L
1L
2R12 = L1R12 − R12L1 , (23)
we come to a generalization of the formula (19).
By introducing an operator
Q : L⊗2
1→ L⊗2
1, Q(L
1L
2) = R−1
12L
1L
2R12
we can present the relation (23) as follows
L1L
2− Q(L
1L
2) = [L
1, L
2]. (24)
It looks like the defining relation of the enveloping algebra of a generalized Lie algebra.
(Though we prefer to use the notations Lk
it is possible to exhibit the maps Q and [ , ] in
120
the basis lj
i⊗ lm
k.) Observe that the map Q is a braiding. Also, note that the operators Sq
and Aq can be expressed in terms of Q and its inverse (cf. [GPS]).
We call the data (g = L1, Q, [ , ]) the gl type quantum (braided) Lie algebra. Note that
if q = 1 (i.e. the symmetry R is involutive) then Q = REnd (V ) and this quantum Lie algebra
is nothing but the generalized Lie algebra gl(VR) and the corresponding mREA becomes
isomorphic to its enveloping algebra.
Let us list the properties of the the quantum Lie algebra in question. The bracket [ , ] is skew-symmetric in the sense of (22). The q-Jacobi identity is valid in the following form
[ , ][ , ]12 = [ , ][ , ]23(I − Q12) . (25) The bracket [ , ] is R-invariant. Essentially, this means that the following relations hold
REnd (V )[ , ]23 = [ , ]12(REnd (V ))23(REnd (V ))12,
REnd (V )[ , ]12 = [ , ]23(REnd (V ))12(REnd (V ))23 . (26)
So, the adjoint action
L1⊲ L
2= [L
1, L
2]
is indeed a representation. By chance (!) the representation ρV ⊗V ∗ coincides with this adjoint
action.
Turn now to the question of the ”sl-reduction”, that is, the passing from the mREA
L(Rq, 1) to the quotient algebra
SL(Rq) := L(Rq, 1)/〈TrRL〉 , TrRL := Tr(CL) , (27)
(see Section 2 for the operator C). The element ℓ := TrRL is central in the mREA, which
can be easily proved by calculating the R-trace in the second space of the matrix relation (8).
To describe the quotient algebra SL(Rq) explicitly, we pass to a new set of generators
f j
i, ℓ, connected with the initial one by a linear transformation:
lj
i= f
j
i+ (Tr(C))−1δ
j
iℓ or L = F + (Tr(C))−1I ℓ , (28)
where F = ‖f j
i‖. Hereafter we assume that TrC = ℓi
i6= 0. (So, the Lie super-algebras
gl(m|m) and their q-deformations are forbidden.) Obviously, TrRF = 0, i.e. the generators
fj
iare dependent.
In terms of the new generators, the commutation relations of the mREA read
R12F1R12F1 − F1R12F1R12 = (eL −ω
Tr(C)ℓ)(R12F1 − F1R12)
ℓ F = F ℓ ,
121
where ω = q − q−1. Now, it is easy to describe the quotient (27) . The matrix F = ‖f j
i‖ of
SL(Rq) generators satisfy the same commutation relations (8) as the matrix L
R12F1R12F1 − F1R12F1R12 = R12F1 − F1R12 , (29)
but the generators fj
iare linearly dependent.
Rewriting this relation in the form similar to (24) we can introduce an sl-type bracket.
However for such a bracket the q-Jacobi identity fails. This is due to the fact the element ℓ
comes in the relations for fj
i(at q = 1 this effect disappears ). Nevertheless, we can construct
a representation
ρV ⊗V ∗ : SL(Rq) → End (V ⊗ V ∗)
which is an analog of the adjoint representation. In order to do so, we rewrite the represen-
tation (20) in terms of the generators fj
iand ℓ. Taking relation (28) into account, we find,
after a short calculation
ρV ⊗V ∗(ℓ) ⊲ ℓ = 0, ρV ⊗V ∗(F1) ⊲ ℓ = 0 ,
ρV ⊗V ∗(ℓ) ⊲ F1 = −ω Tr(C)F1
ρV ⊗V ∗(F1) ⊲ F
2= F1R12 − R12F1 + ωR12F1R
−1
12. (30)
Namely, the last formula from this list defines the representation ρV ⊗V ∗ . However, in
contrast with the mREA L(Rq, 1), this map is different from that defined by the bracket [ , ]
reduced to the space span (f j
i). This is reason why the ”q-adjoint” representation cannot be
presented in the form (25). (Also, note that though ℓ is central it acts in a non-trivial way
on the elements fj
i.)
Moreover, any object U of the category SW(V) above such that
ρU(ℓ) = χ IU , χ ∈ K
is a scalar operator, can be equipped with an SL(Rq)-module structure. First, let us observe
that for any representation ρU : L(Rq, 1) → End (U) and for any z ∈ K the map
ρz
U: L(Rq, 1) → End (U), ρz
U(lj
i) = zρU (lj
i) + δ
j
i(1 − z)(q − q−1)−1 IU
is a representation of this algebra as well.
By using this freedom we can convert a given representation ρU : L(Rq, 1) → End (U)
with the above property into that ρz
Usuch that ρz
U(ℓ) = 0. Thus we get a representation of
the algebra SL(Rq). Explicitly, this representation is given by the formula
ρ(f j
i) =
1
ξ
(
ρ(lji) − (Tr(C))−1ρ(ℓ) δ
j
i
)
, ξ = 1 − (q − q−1)(Tr(C))−1χ . (31)
The data (span (f j
i), Q, [ , ]) where the bracket stands for the l.h.s. of (24) restricted to
span (f j
i) is called the sl-type quantum (braided) Lie algebra.
122
Note that in the particular case related to the QG Uq(sl(n)) this quantum algebra can
be treated in terms of [LS] where an axiomatic approach to the corresponding Lie algebra-
like object is given. However, we think that any general axiomatic definition of such objects
is somewhat useless (unless the corresponding symmetry is involutive). Our viewpoint is
motivated by the fact that for Bn, Cn, Dn series there do not exist ”quantum Lie algebras”
such that their enveloping algebras have good deformation properties. As for the An series
(or more generally, for any skew-invertible Hecke symmetry) such objects exist and can be
explicitly exhibited via the mREA. Their properties differ from those listed in [W, GM] in
the framework of an axiomatic approach to Lie algebra-like objects.
Completing the paper, we want to emphasize that the above coproduct can be useful for
definition of a ”braided (co)adjoint vector field”. In the L(Rq, 1) case these fields are naturally
introduced through the above adjoint action extended to the symmetric algebra of the space
L1 by means of this coproduct. The symmetric algebra can be defined via the above operators
Sq and Aq. In the SL(Rq) case a similar treatment is possible if Tr C 6= 0.
References
[CP] Chari V., Pressley. A. A guide to quantum groups, Cambridge University Press, Cam-
bridge, 1994.
[DGG] Delius G.W., Gardner C., Gould M.D. The structure of quantum Lie algebra for the
classical series Bl, Cl and Dl, J.Phys.A: Math. Gen. 31 (1998) 1995–2019.
[GM] Gomez X., Majid S. Braided Lie algebras and bicovariant differential calculi over co-
quasitriangular Hopf algebras, J. of Algebra, 261 (2003) 334–388.
[G1] Gurevich D. Generalized translation operators on Lie groups, Engl. transl.: Soviet J.
Contemporary Math. Anal. 18 (1983) 57–90.
[G2] Gurevich D. Algebraic aspects of the Yang-Baxter equation, English translation:
Leningrad Math. 2 (1991) 801 – 828.
[GPS] Gurevich D., Pyatov P., Saponov P. Representations theory of (modified) Reflection
Equation Algebra of GL(m|n) type, St. Petersburg Math. J., submitted.
[LS] Lyubashenko V., Sudbery A. Generalized Lie algebras of type An, J. Math. Phys. 39
(1998) 3487-3504.
[PP] Polishchuk A., Positselski L. Quadratic Algebras, University Lecture Series, 37, Amer-
ican Mathematical Society, Providence, Rhode Island.
[Sh] Scheunert M. Generalized Lie algebra, J. Math. Phys. 20 (1979) 712-720.
123
[W] Woronowicz S. Differential Calculus on Compact Matrix Pseudogroups (Quantum
groups), CMP 122 (1989) 125–170.
124
On the automorphism groups of q-enveloping
algebras of nilpotent Lie algebras
Stephane Launois∗
Abstract
We investigate the automorphism group of the quantised enveloping algebra U+q (g)
of the positive nilpotent part of certain simple complex Lie algebras g in the case where
the deformation parameter q ∈ C∗ is not a root of unity. Studying its action on the
set of minimal primitive ideals of U+q (g) we compute this group in the cases where g =
sl3 and g = so5 confirming a Conjecture of Andruskiewitsch and Dumas regarding the
automorphism group of U+q (g). In the case where g = sl3, we retrieve the description of the
automorphism group of the quantum Heisenberg algebra that was obtained independently
by Alev and Dumas, and Caldero. In the case where g = so5, the automorphism group of
U+q (g) was computed in [16] by using previous results of Andruskiewitsch and Dumas. In
this paper, we give a new (simpler) proof of the Conjecture of Andruskiewitsch and Dumas
in the case where g = so5 based both on the original proof and on graded arguments
developed in [17] and [18].
Introduction
In the classical situation, there are few results about the automorphism group of the envelop-
ing algebra U(L) of a Lie algebra L over C; except when dimL ≤ 2, these groups are known
to possess “wild” automorphisms and are far from being understood. For instance, this is the
case when L is the three-dimensional abelian Lie algebra [22], when L = sl2 [14] and when L
is the three-dimensional Heisenberg Lie algebra [1].
In this paper we study the quantum situation. More precisely, we study the automorphism
group of the quantised enveloping algebra U+q
(g) of the positive nilpotent part of a finite
dimensional simple complex Lie algebra g in the case where the deformation parameter q ∈ C∗
is not a root of unity. Although it is a common belief that quantum algebras are ”rigid” and
so should possess few symmetries, little is known about the automorphism group of U+
q(g).
∗School of Mathematics, University of Edinburgh, James Clerk Maxwell Building, King’s Buildings, May-
field Road, Edinburgh EH9 3JZ, Scotland. E-mail: [email protected]. Current address (From
01 August 2007): Institute of Mathematics, Statistics and Actuarial Science, University of Kent at Canter-
bury, CT2 7NF, UK. This research was supported by a Marie Curie Intra-European Fellowship within the 6th
European Community Framework Programme.
125
Indeed, until recently, this group was known only in the case where g = sl3 whereas the
structure of the automorphism group of the augmented form Uq(b+), where b
+ is the positive
Borel subalgebra of g, has been described in [9] in the general case.
The automorphism group of U+q
(sl3) was computed independently by Alev-Dumas, [2],
and Caldero, [8], who showed that
Aut(U+
q(sl3)) ≃ (C∗)2 ⋊ S2.
Recently, Andruskiewitsch and Dumas, [4] have obtained partial results on the automorphism
group of U+q
(so5). In view of their results and the description of Aut(U+q
(sl3)), they have
proposed the following conjecture.
Conjecture (Andruskiewitsch-Dumas, [4, Problem 1]):
Aut(U+
q(g)) ≃ (C∗)rk(g)
⋊ autdiagr(g),
where autdiagr(g) denotes the group of automorphisms of the Dynkin diagram of g.
Recently we proved this conjecture in the case where g = so5, [16], and, in collaboration
with Samuel Lopes, in the case where g = sl4, [18]. The techniques in these two cases are
very different. Our aim in this paper is to show how one can prove the Andruskiewitsch-
Dumas Conjecture in the cases where g = sl3 and g = so5 by first studying the action of
Aut(U+
q(g)) on the set of minimal primitive ideals of U+
q(g) - this was the main idea in [16]
-, and then using graded arguments as developed in [17] and [18]. This strategy leads us to
a new (simpler) proof of the Andruskiewitsch-Dumas Conjecture in the case where g = so5.
Throughout this paper, N denotes the set of nonnegative integers, C∗ := C \ 0 and q is
a nonzero complex number that is not a root of unity.
1 Preliminaries
In this section, we present the H-stratification theory of Goodearl and Letzter for the positive
part U+q
(g) of the quantised enveloping algebra of a simple finite-dimensional complex Lie
algebra g. In particular, we present a criterion (due to Goodearl and Letzter) that charac-
terises the primitive ideals of U+q
(g) among its prime ideals. In the next section, we will use
this criterion in order to describe the primitive spectrum of U+
q(g) in the cases where g = sl3
and g = so5.
1.1 Quantised enveloping algebras and their positive parts.
Let g be a simple Lie C-algebra of rank n. We denote by π = α1, . . . , αn the set of simple
roots associated to a triangular decomposition g = n− ⊕ h ⊕ n
+. Recall that π is a basis of
126
an euclidean vector space E over R, whose inner product is denoted by ( , ) (E is usually
denoted by h∗R
in Bourbaki). We denote by W the Weyl group of g, that is, the subgroup
of the orthogonal group of E generated by the reflections si := sαi, for i ∈ 1, . . . , n, with
reflecting hyperplanes Hi := β ∈ E | (β, αi) = 0, i ∈ 1, . . . , n. The length of w ∈ W
is denoted by l(w). Further, we denote by w0 the longest element of W . We denote by
R+ the set of positive roots and by R the set of roots. Set Q+ := Nα1 ⊕ · · · ⊕ Nαn and
Q := Zα1⊕· · ·⊕Zαn. Finally, we denote by A = (aij) ∈ Mn(Z) the Cartan matrix associated
to these data. As g is simple, aij ∈ 0,−1,−2,−3 for all i 6= j.
Recall that the scalar product of two roots (α, β) is always an integer. As in [5], we assume
that the short roots have length√
2.
For all i ∈ 1, . . . , n, set qi := q(αi,αi)
2 and
[
m
k
]
i
:=(qi − q−1
i) . . . (qm−1
i− q1−m
i)(qm
i− q−m
i)
(qi − q−1
i) . . . (qk
i− q−k
i)(qi − q−1
i) . . . (qm−k
i− qk−m
i)
for all integers 0 ≤ k ≤ m. By convention,
[
m
0
]
i
:= 1.
The quantised enveloping algebra Uq(g) of g over C associated to the previous data is the
C-algebra generated by the indeterminates E1, . . . , En, F1, . . . , Fn,K±1
1, . . . ,K±1
nsubject to
the following relations:
KiKj = KjKi
KiEjK−1
i= q
aij
iEj and KiFjK
−1
i= q
−aij
iFj
EiFj − FjEi = δij
Ki − K−1
i
qi − q−1
i
and the quantum Serre relations:
1−aij∑
k=0
(−1)k
[
1 − aij
k
]
i
E1−aij−k
iEjE
k
i= 0 (i 6= j) (1)
and1−aij∑
k=0
(−1)k
[
1 − aij
k
]
i
F1−aij−k
iFjF
k
i= 0 (i 6= j).
We refer the reader to [5, 13, 15] for more details on this (Hopf) algebra. Further, as
usual, we denote by U+q
(g) (resp. U−q
(g)) the subalgebra of Uq(g) generated by E1, . . . , En
(resp. F1, . . . , Fn) and by U0 the subalgebra of Uq(g) generated by K±1
1, . . . ,K±1
n. Moreover,
for all α = a1α1 + · · · + anαn ∈ Q, we set
Kα := Ka1
1· · ·Kan
n.
127
As in the classical case, there is a triangular decomposition as vector spaces:
U−q
(g) ⊗ U0 ⊗ U+
q(g) ≃ Uq(g).
In this paper we are concerned with the algebra U+q
(g) that admits the following presentation,
see [13, Theorem 4.21]. The algebra U+
q(g) is (isomorphic to) the C-algebra generated by n
indeterminates E1, . . . , En subject to the quantum Serre relations (1).
1.2 PBW-basis of U+
q(g).
To each reduced decomposition of the longest element w0 of the Weyl group W of g, Lusztig
has associated a PBW basis of U+q
(g), see for instance [19, Chapter 37], [13, Chapter 8] or [5,
I.6.7]. The construction relates to a braid group action by automorphisms on U+
q(g). Let us
first recall this action. For all s ∈ N and i ∈ 1, . . . , n, we set
[s]i :=qs
i− q−s
i
qi − q−1
i
and [s]i! := [1]i . . . [s − 1]i[s]i.
As in [5, I.6.7], we denote by Ti, for 1 ≤ i ≤ n, the automorphism of U+q
(g) defined by:
Ti(Ei) = −FiKi,
Ti(Ej) =
−aij∑
s=0
(−1)s−aijq−s
iE
(−aij−s)
iEjE
(s)
i, i 6= j
Ti(Fi) = −K−1
iEi,
Ti(Fj) =
−aij∑
s=0
(−1)s−aij qs
iF
(s)
iFjF
(−aij−s)
i, i 6= j
Ti(Kα) = Ksi(α), α ∈ Q,
where E(s)
i:=
Esi
[s]i!and F
(s)
i:=
Fsi
[s]i!for all s ∈ N. It was proved by Lusztig that the automor-
phisms Ti satisfy the braid relations, that is, if sisj has order m in W , then
TiTjTi · · · = TjTiTj . . . ,
where there are exactly m factors on each side of this equality.
The automorphisms Ti can be used in order to describe PBW bases of U+q
(g) as fol-
lows. It is well-known that the length of w0 is equal to the number N of positive roots
of g. Let si1· · · siN
be a reduced decomposition of w0. For k ∈ 1, . . . , N, we set βk :=
si1· · · sik−1
(αik). Then β1, . . . , βN is exactly the set of positive roots of g. Similarly, we
define elements Eβkof Uq(g) by
Eβk:= Ti1
· · ·Tik−1(Eik
).
Note that the elements Eβkdepend on the reduced decomposition of w0. The following
well-known results were proved by Lusztig and Levendorskii-Soibelman.
128
Theorem 1.1 (Lusztig and Levendorskii-Soibelman).
1. For all k ∈ 1, . . . , N, the element Eβkbelongs to U+
q(g).
2. If βk = αi, then Eβk= Ei.
3. The monomials Ek1
β1· · ·EkN
βN, with k1, . . . , kN ∈ N, form a linear basis of U+
q(g).
4. For all 1 ≤ i < j ≤ N , we have
EβjEβi
− q−(βi,βj)EβiEβj
=∑
aki+1,...,kj−1E
ki+1
βi+1· · ·E
kj−1
βj−1,
where each aki+1,...,kj−1belongs to C.
As a consequence of this result, U+
q(g) can be presented as a skew-polynomial algebra:
U+
q(g) = C[Eβ1
][Eβ2;σ2, δ2] · · · [EβN
;σN , δN ],
where each σi is a linear automorphism and each δi is a σi-derivation of the appropriate
subalgebra. In particular, U+
q(g) is a noetherian domain and its group of invertible elements
is reduced to nonzero complex numbers.
1.3 Prime and primitive spectra of U+
q(g).
We denote by Spec(U+q
(g)) the set of prime ideals of U+q
(g). First, as q is not a root of unity,
it was proved by Ringel [21] (see also [10, Theorem 2.3]) that, as in the classical situation,
every prime ideal of U+q
(g) is completely prime.
In order to study the prime and primitive spectra of U+
q(g), we will use the stratification
theory developed by Goodearl and Letzter. This theory allows the construction of a partition
of these two sets by using the action of a suitable torus on U+
q(g). More precisely, the torus
H := (C∗)n acts naturally by automorphisms on U+q
(g) via:
(h1, . . . , hn).Ei = hiEi for all i ∈ 1, . . . , n.
(It is easy to check that the quantum Serre relations are preserved by the group H.) Recall
(see [4, 3.4.1]) that this action is rational. (We refer the reader to [5, II.2.] for the defintion of
a rational action.) A non-zero element x of U+
q(g) is an H-eigenvector of U+
q(g) if h.x ∈ C
∗x
for all h ∈ H. An ideal I of U+q
(g) is H-invariant if h.I = I for all h ∈ H. We denote by H-
Spec(U+
q(g)) the set of all H-invariant prime ideals of U+
q(g). It turns out that this is a finite
set by a theorem of Goodearl and Letzter about iterated Ore extensions, see [11, Proposition
4.2]. In fact, one can be even more precise in our situation. Indeed, in [12], Gorelik has
also constructed a stratification of the prime spectrum of U+q
(g) using tools coming from
representation theory. It turns out that her stratification coincides with the H-stratification,
so that we deduce from [12, Corollary 7.1.2] that
129
Proposition 1.2 (Gorelik). U+q
(g) has exactly |W | H-invariant prime ideals.
The action of H on U+
q(g) allows via the H-stratification theory of Goodearl and Letzter
(see [5, II.2]) the construction of a partition of Spec(U+q
(g)) as follows. If J is an H-invariant
prime ideal of U+
q(g), we denote by Spec
J(U+
q(g)) the H-stratum of Spec(U+
q(g)) associated
to J . Recall that SpecJ(U+q
(g)) := P ∈ Spec(U+q
(g)) |⋂
h∈H h.P = J. Then the H-strata
SpecJ(U+
q(g)) (J ∈ H-Spec(U+
q(g))) form a partition of Spec(U+
q(g)) (see [5, II.2]):
Spec(U+
q(g)) =
⊔
J∈H-Spec(U+q (g))
SpecJ(U+
q(g)).
Naturally, this partition induces a partition of the set Prim(U+
q(g)) of all (left) primitive ideals
of U+q
(g) as follows. For all J ∈ H-Spec(U+q
(g)), we set PrimJ(U+q
(g)) := SpecJ(U+q
(g)) ∩
Prim(U+
q(g)). Then it is obvious that the H-strata PrimJ(U+
q(g)) (J ∈ H-Spec(U+
q(g))) form
a partition of Prim(U+q
(g)):
Prim(U+
q(g)) =
⊔
J∈H-Spec(U+q (g))
PrimJ(U+
q(g)).
More interestingly, because of the finiteness of the set of H-invariant prime ideals of U+q
(g), the
H-stratification theory provides a useful tool to recognise primitive ideals without having to
find all its irreductible representations! Indeed, following previous works of Hodges-Levasseur,
Joseph, and Brown-Goodearl, Goodearl and Letzter have characterised the primitive ideals
of U+q
(g) as follows, see [11, Corollary 2.7] or [5, Theorem II.8.4].
Theorem 1.3 (Goodearl-Letzter). PrimJ(U+
q(g)) (J ∈ H-Spec(U+
q(g))) coincides with those
primes in SpecJ(U+q
(g)) that are maximal in SpecJ(U+q
(g)).
2 Automorphism group of U+q (g)
In this section, we investigate the automorphism group of U+
q(g) viewed as the algebra gen-
erated by n indeterminates E1, . . . , En subject to the quantum Serre relations. This algebra
has some well-identified automorphisms. First, there are the so-called torus automorphisms;
let H = (C∗)n, where n still denotes the rank of g. As U+q
(g) is the C-algebra generated
by n indeterminates subject to the quantum Serre relations, it is easy to check that each
λ = (λ1, . . . , , λn) ∈ H determines an algebra automorphism φλ
of U+q
(g) with φλ(Ei) = λiEi
for i ∈ 1, . . . , n, with inverse φ−1
λ= φ
λ−1 . Next, there are the so-called diagram auto-
morphisms coming from the symmetries of the Dynkin diagram of g. Namely, let w be an
automorphism of the Dynkin diagram of g, that is, w is an element of the symmetric group Sn
such that (αi, αj) = (αw(i), αw(j)) for all i, j ∈ 1, . . . , n. Then one defines an automorphism,
also denoted w, of U+
q(g) by: w(Ei) = Ew(i). Observe that
φλ w = w φ(λw(1) ,...,,λw(n))
.
130
We denote by G the subgroup of Aut(U+q
(g)) generated by the torus automorphisms and
the diagram automorphisms. Observe that
G ≃ H ⋊ autdiagr(g),
where autdiagr(g) denotes the set of diagram automorphisms of g.
The group Aut(U+
q(sl3)) was computed independently by Alev and Dumas, see [2, Propo-
sition 2.3] , and Caldero, see [8, Proposition 4.4]; their results show that, in the case where
g = sl3, we have
Aut(U+
q(sl3)) = G.
About ten years later, Andruskiewitsch and Dumas investigated the case where g = so5, see
[4]. In this case, they obtained partial results that lead them to the following conjecture.
Conjecture (Andruskiewitsch-Dumas, [4, Problem 1]):
Aut(U+
q(g)) = G.
This conjecture was recently confirmed in two new cases: g = so5, [16], and g = sl4, [18].
Our aim in this section is to show how one can use the action of the automorphism group of
U+q
(g) on the primitive spectrum of this algebra in order to prove the Andruskiewitsch-Dumas
Conjecture in the cases where g = sl3 and g = so5.
2.1 Normal elements of U+
q(g)
Recall that an element a of U+
q(g) is normal provided the left and right ideals generated by
a in U+q
(g) coincide, that is, if
aU+
q(g) = U+
q(g)a.
In the sequel, we will use several times the following well-known result concerning normal
elements of U+
q(g).
Lemma 2.1. Let u and v be two nonzero normal elements of U+q
(g) such that 〈u〉 = 〈v〉.
Then there exist λ, µ ∈ C∗ such that u = λv and v = µu.
Proof. It is obvious that units λ, µ exist with these properties. However, the set of units of
U+
q(g) is precisely C
∗.
2.2 N-grading on U+
q(g) and automorphisms
As the quantum Serre relations are homogeneous in the given generators, there is an N-grading
on U+
q(g) obtained by assigning to Ei degree 1. Let
U+
q(g) =
⊕
i∈N
U+
q(g)i (2)
131
be the corresponding decomposition, with U+q
(g)i the subspace of homogeneous elements of
degree i. In particular, U+q
(g)0 = C and U+q
(g)1 is the n-dimensional space spanned by the
generators E1, . . . , En. For t ∈ N set U+
q(g)≥t =
⊕
i≥tU+
q(g)i and define U+
q(g)≤t similarly.
We say that the nonzero element u ∈ U+q
(g) has degree t, and write deg(u) = t, if
u ∈ U+
q(g)≤t \ U+
q(g)≤t−1 (using the convention that U+
q(g)≤−1 = 0). As U+
q(g) is a
domain, deg(uv) = deg(u) + deg(v) for u, v 6= 0.
Definition 2.2. Let A =⊕
i∈NAi be an N-graded C-algebra with A0 = C which is generated
as an algebra by A1 = Cx1 ⊕ · · · ⊕Cxn. If for each i ∈ 1, . . . , n there exist 0 6= a ∈ A and a
scalar qi,a 6= 1 such that xia = qi,aaxi, then we say that A is an N-graded algebra with enough
q-commutation relations.
The algebra U+q
(g), endowed with the grading just defined, is a connected N-graded al-
gebra with enough q-commutation relations. Indeed, if i ∈ 1, . . . , n, then there exists
u ∈ U+q
(g) such that Eiu = q•uEi where • is a nonzero integer. This can be proved as
follows. As g is simple, there exists an index j ∈ 1, . . . , n such that j 6= i and aij 6= 0,
that is, aij ∈ −1,−2,−3. Then sisj is a reduced expression in W , so that one can find a
reduced expression of w0 starting with sisj, that is, one can write
w0 = sisjsi3. . . siN
.
With respect to this reduced expression of w0, we have with the notation of Section 1.2:
β1 = αi and β2 = si(αj) = αj − aijαi
Then it follows from Theorem 1.1 that Eβ1= Ei, Eβ2
= Eαj−aijαiand
EiEβ2= q(αi,αj−aijαi)Eβ2
Ei,
that is,
EiEβ2= q−(αi,αj)Eβ2
Ei.
As aij 6= 0, we have (αi, αj) 6= 0 and so q−(αi,αj) 6= 1 since q is not a root of unity. So we
have just proved:
Proposition 2.3. U+q
(g) is a connected N-graded algebra with enough q-commutation rela-
tions.
One of the advantages of N-graded algebras with enough q-commutation relations is
that any automorphism of such an algebra must conserve the valuation associated to the
N-graduation. More precisely, as U+q
(g) is a connected N-graded algebra with enough q-
commutation relations, we deduce from [18] (see also [17, Proposition 3.2]) the following
result.
Corollary 2.4. Let σ ∈ Aut(U+
q(g)) and x ∈ U+
q(g)d \ 0. Then σ(x) = yd + y>d, for some
yd ∈ U+q
(g)d \ 0 and y>d ∈ U+q
(g)≥d+1.
132
2.3 The case where g = sl3
In this section, we investigate the automorphism group of U+q
(g) in the case where g = sl3. In
this case the Cartan matrix is A =
(
2 −1
−1 2
)
, so that U+q
(sl3) is the C-algebra generated
by two indeterminates E1 and E2 subject to the following relations:
E2
1E2 − (q + q−1)E1E2E1 + E2E2
1 = 0 (3)
E2
2E1 − (q + q−1)E2E1E2 + E1E
2
2= 0 (4)
We often refer to this algebra as the quantum Heisenberg algebra, and sometimes we denote
it by H, as in the classical situation the enveloping algebra of sl+
3is the so-called Heisenberg
algebra.
We now make explicit a PBW basis of H. The Weyl group of sl3 is isomorphic to the
symmetric group S3, where s1 is identified with the transposition (1 2) and s2 is identified
with (2 3). Its longest element is then w0 = (13); it has two reduced decompositions: w0 =
s1s2s1 = s2s1s2. Let us choose the reduced decomposition s1s2s1 of w0 in order to construct
a PBW basis of U+
q(sl3). According to Section 1.2, this reduced decomposition leads to the
following root vectors:
Eα1= E1, Eα1+α2
= T1(E2) = −E1E2 + q−1E2E1 and Eα2= T1T2(E1) = E2.
In order to simplify the notation, we set E3 := −E1E2 + q−1E2E1. Then, it follows from
Theorem 1.1 that The monomials Ek1
1E
k3
3E
k2
2, with k1, k2, k3 nonnegative integers, form a PBW-basis of
U+
q(sl3). H is the iterated Ore extension over C generated by the indeterminates E1, E3, E2
subject to the following relations:
E3E1 = q−1E1E3, E2E3 = q−1E3E2, E2E1 = qE1E2 + qE3.
In particular, H is a Noetherian domain, and its group of invertible elements is reduced
to C∗. It follows from the previous commutation relations between the root vectors that E3 is
a normal element in H, that is, E3H = HE3.
In order to describe the prime and primitive spectra of H, we need to introduce two other
elements. The first one is the root vector E′3
:= T2(E1) = −E2E1 + q−1E1E2. This root
vector would have appeared if we have choosen the reduced decomposition s2s1s2 of w0 in
order to construct a PBW basis of H. It follows from Theorem 1.1 that E′3
q-commutes with
E1 and E2, so that E′3
is also a normal element of H. Moreover, one can describe the centre
133
of H using the two normal elements E3 and E′3. Indeed, in [3, Corollaire 2.16], Alev and
Dumas have described the centre of U+q
(sln); independently Caldero has described the centre
of U+
q(g) for arbitrary g, see [7]. In our particular situation, their results show that the centre
Z(H) of H is a polynomial ring in one variable Z(H) = C[Ω], where Ω = E3E′3.
We are now in position to describe the prime and primitive spectra of H = U+
q(sl(3));
this was first achieved by Malliavin who obtained the following picture for the poset of prime
ideals of H, see [20, Theoreme 2.4]:
〈〈E1, E2 − β〉〉
AA
AA
AA
AA
AA
AA
AA
AA
〈〈E1, E2〉〉
<<
<<
<<
<<
<<
<<
<<
<
〈〈E1 − α,E2〉〉
〈E1〉
~~~~~~~~~~~~~~~~
TTT
TTTT
TTTT
TTTTT
TTTT
TTTT
TTTT
TTTT
TTTT
T〈E2〉
jjjj
jjjj
jjjj
jjjjj
jjjj
jjjj
jjjj
jjjj
jjjj
AA
AA
AA
AA
AA
AA
AA
AA
〈〈E3〉〉
OO
OO
OO
OO
OO
OO
OO
OO
OO
OO
OO
OO
OO
〈〈Ω − γ〉〉 〈〈E′3〉〉
oooooooooooooooooooooooooo
〈0〉
where α, β, γ ∈ C∗.
Recall from Section 1.3 that the torus H = (C∗)2 acts on U+
q(sl3) by automorphisms and
that the H-stratification theory of Goodearl and Letzter constructs a partition of the prime
spectrum of U+
q(sl3) into so-called H-strata, this partition being indexed by the H-invariant
prime ideals of U+q
(sl3). Using this description of Spec(U+q
(sl3)), it is easy to identify the
6 = |W | H-invariant prime ideals of H and their corresponding H-strata. As E1, E2, E3 and
E′3
are H-eigenvectors, the 6 H-invariant primes are:
〈0〉, 〈E3〉, 〈E′3〉, 〈E1〉, 〈E2〉 and 〈E1, E2〉.
Moreover the corresponding H-strata are:
Spec〈0〉(H) = 〈0〉 ∪ 〈Ω − γ〉 | γ ∈ C∗,
Spec〈E3〉(H) = 〈E3〉,
Spec〈E′3〉(H) = 〈E′
3〉,
Spec〈E1〉(H) = 〈E1〉 ∪ 〈E1, E2 − β〉 | β ∈ C
∗,
Spec〈E2〉(H) = 〈E2〉 ∪ 〈E1 − α,E2〉 | α ∈ C
∗
and Spec〈E1,E2〉(H) = 〈E1, E2〉.
134
We deduce from this description of the H-strata and the the fact that primitive ideals
are exactly those primes that are maximal within their H-strata, see Theorem 1.3, that the
primitive ideals of U+
q(sl3) are exactly those primes that appear in double brackets in the
previous picture.
We now investigate the group of automorphisms of H = U+
q(sl3). In that case, the torus
acting naturally on U+q
(sl3) is H = (C∗)2, there is only one non-trivial diagram automorphism
w that exchanges E1 and E2, and so the subgroup G of Aut(U+
q(sl3)) generated by the torus
and diagram automorphisms is isomorphic to the semi-direct product (C∗)2 ⋊ S2. We want
to prove that Aut(U+
q(sl3)) = G.
In order to do this, we study the action of Aut(U+q
(sl3)) on the set of primitive ideals that
are not maximal. As there are only two of them, 〈E3〉 and 〈E′3〉, an automorphism of H will
either fix them or permute them.
Let σ be an automorphism of U+
q(sl3). It follows from the previous observation that
either σ(〈E3〉) = 〈E3〉 and σ(〈E′3〉) = 〈E′
3〉,
or σ(〈E3〉) = 〈E′3〉 and σ(〈E′
3〉) = 〈E3〉.
As it is clear that the diagram automorphism w permutes the ideals 〈E3〉 and 〈E′3〉, we get
that there exists an automorphism g ∈ G such that
g σ(〈E3〉) = 〈E3〉 and g σ(〈E′3〉) = 〈E′
3〉.
Then, as E3 and E′3
are normal, we deduce from Lemma 2.1 that there exist λ, λ′ ∈ C∗ such
that
g σ(E3) = λE3 and g σ(E′3) = λ′E′
3.
In order to prove that g σ is an element of G, we now use the N-graduation of U+
q(sl3)
introduced in Section 2.2. With respect to this graduation, E1 and E2 are homogeneous of
degree 1, and so E3 and E′3
are homogeneous of degree 2. Moreover, as (q−2 − 1)E1E2 =
E3 + q−1E′3, we deduce from the above discussion that
g σ(E1E2) =1
q−2 − 1
(λE3 + q−1λ′E′
3
)
has degree two. On the other hand, as U+
q(sl3) is a connected N-graded algebra with enough
q-commutation relations by Proposition 2.3, it follows from Corollary 2.4 that σ(E1) = a1E1+
a2E2 + u and σ(E2) = b1E1 + b2E2 + v, where (a1, a2), (b1, b2) ∈ C2 \ (0, 0), and u, v ∈
U+q
(sl3) are linear combinations of homogeneous elements of degree greater than one. As
g σ(E1).g σ(E2) has degree two, it is clear that u = v = 0. To conclude that g σ ∈ G,
it just remains to prove that a2 = 0 = b1. This can be easily shown by using the fact that
gσ(−E1E2 +q−1E2E1) = gσ(E3) = λE3; replacing gσ(E1) and gσ(E2) by a1E1 +a2E2
and b1E1 + b2E2 respectively, and then identifying the coefficients in the PBW basis, leads to
135
a2 = 0 = b1, as required. Hence we have just proved that g σ ∈ G, so that σ itself belongs to
G the subgroup of Aut(U+q
(sl3)) generated by the torus and diagram automorphisms. Hence
one can state the following result that confirms the Andruskiewitsch-Dumas Conjecture.
Proposition 2.5. Aut(U+q
(sl3)) ≃ (C∗)2 ⋊ autdiagr(sl3)
This result was first obtained independently by Alev and Dumas, [2, Proposition 2.3],
and Caldero, [8, Proposition 4.4], but using somehow different methods; they studied this
automorphism group by looking at its action on the set of normal elements of U+q
(sl3).
2.4 The case where g = so5
In this section we investigate the automorphism group of U+q
(g) in the case where g =
so5. In this case there are no diagram automorphisms, so that the Andruskiewitsch-Dumas
Conjecture asks whether every automorphism of U+q
(so5) is a torus automorphism. In [16]
we have proved their conjecture when g = so5. The aim of this section is to present a slightly
different proof based both on the original proof and on the recent proof by S.A. Lopes and
the author of the Andruskiewitsch-Dumas Conjecture in the case where g is of type A3.
In the case where g = so5, the Cartan matrix is A =
(
2 −2
−1 2
)
, so that U+
q(so5) is
the C-algebra generated by two indeterminates E1 and E2 subject to the following relations:
E3
1E2 − (q2 + 1 + q−2)E2
1E2E1 + (q2 + 1 + q−2)E1E2E
2
1+ E2E
3
1= 0 (5)
E2
2E1 − (q2 + q−2)E2E1E2 + E1E2
2 = 0 (6)
We now make explicit a PBW basis of U+
q(so5). The Weyl group of so5 is isomorphic to
the dihedral group D(4). Its longest element is w0 = −id; it has two reduced decompositions:
w0 = s1s2s1s2 = s2s1s2s1. Let us choose the reduced decomposition s1s2s1s2 of w0 in order
to construct a PBW basis of U+q
(so5). According to Section 1.2, this reduced decomposition
leads to the following root vectors:
Eα1= E1, E2α1+α2
= T1(E2) =1
(q + q−1)
(E2
1E2 − q−1(q + q−1)E1E2E1 + q−2E2E2
1
),
Eα1+α2= T1T2(E1) = −E1E2 + q−2E2E1 and Eα2
= T1T2T1(E2) = E2.
In order to simplify the notation, we set E3 := −Eα1+α2and E4 := E2α1+α2
. Then, it
follows from Theorem 1.1 that The monomials Ek1
1E
k4
4E
k3
3E
k2
2, with k1, k2, k3, k4 nonnegative integers, form a PBW-
basis of U+
q(so5).
136
U+q
(so5) is the iterated Ore extension over C generated by the indeterminates E1, E4,
E3, E2 subject to the following relations:
E4E1 = q−2E1E4
E3E1 = E1E3 − (q + q−1)E4, E3E4 = q−2E4E3,
E2E1 = q2E1E2 − q2E3, E2E4 = E4E2 −q2−1
q+q−1 E2
3, E2E3 = q−2E3E2.
In particular, U+q
(so5) is a Noetherian domain, and its group of invertible elements is
reduced to C∗.
Before describing the automorphism group of U+
q(so5), we first describe the centre and the
primitive ideals of U+q
(so5). The centre of U+q
(g) has been described in general by Caldero,
[7]. In the case where g = so5, his result shows that Z(U+
q(so5)) is a polynomial algebra in
two indeterminates
Z(U+
q(so5)) = C[z, z′],
where
z = (1 − q2)E1E3 + q2(q + q−1)E4
and
z′ = −(q2 − q−2)(q + q−1)E4E2 + q2(q2 − 1)E2
3.
Recall from Section 1.3 that the torus H = (C∗)2 acts on U+q
(so5) by automorphisms and
that the H-stratification theory of Goodearl and Letzter constructs a partition of the prime
spectrum of U+q
(so5) into so-called H-strata, this partition being indexed by the 8 = |W |
H-invariant prime ideals of U+
q(so5). In [16], we have described these eight H-strata. More
precisely, we have obtained the following picture for the poset Spec(U+q
(so5)),
137
〈〈E1, E2 − β〉〉
BB
BB
BB
BB
BB
BB
BB
BB
B
〈〈E1, E2〉〉
CC
CC
CC
CC
CC
CC
CC
CC
C
〈〈E1 − α,E2〉〉
|||||||||||||||||
〈E1〉
Q〈E2〉
mmmmmmmmmmmmmmmmmmmmmmmmmmmmm
〈〈z, z′ − δ〉〉
BB
BB
BB
BB
BB
BB
BB
BB
B
〈〈E3〉〉
〈〈z − γ, z′ − δ〉〉 〈〈E′3〉〉
mmmmmmmmmmmmmmmmmmmmmmmmmmmmmm
〈〈z − γ, z′〉〉
|||||||||||||||||
〈z〉
CC
CC
CC
CC
CC
CC
CC
CC
C
I 〈z′〉
〈0〉
where α, β, γ, δ ∈ C∗, E′
3:= E1E2 − q2E2E1 and
I = 〈P (z, z′)〉 | P is a unitary irreductible polynomial of C[z, z′], P 6= z, z′.
As the primitive ideals are those primes that are maximal in their H-strata, see Theorem
1.3, we deduced from this description of the prime spectrum that the primitive ideals of
U+
q(so5) are the following:
• 〈z − α, z′ − β〉 with (α, β) ∈ C2 \ (0, 0).
• 〈E3〉 and 〈E′3〉.
• 〈E1 − α,E2 − β〉 with (α, β) ∈ C2 such that αβ = 0.
(They correspond to the “double brackets” prime ideals in the above picture.)
Among them, two only are not maximal, 〈E3〉 and 〈E′3〉. Unfortunately, as E3 and E′
3are
not normal in U+
q(so5), one cannot easily obtain information using the fact that any automor-
phism of U+q
(so5) will either preserve or exchange these two prime ideals. Rather than using
this observation, we will use the action of Aut(U+
q(so5)) on the set of maximal ideals of height
two. Because of the previous description of the primitive spectrum of U+q
(so5), the height two
maximal ideals in U+
q(so5) are those 〈z − α, z′ − β〉 with (α, β) ∈ C
2\(0, 0). In [16, Proposi-
tion 3.6], we have proved that the group of units of the factor algebra U+q
(so5)/〈z − α, z′ − β〉
is reduced to C∗ if and only if both α and β are nonzero. Consequently, if σ is an automor-
138
phism of U+q
(so5) and α ∈ C∗, we get that:
σ(〈z − α, z′〉) = 〈z − α′, z′〉 or 〈z, z′ − β′〉,
where α′, β′ ∈ C∗. Similarly, if σ is an automorphism of U+
q(so5) and β ∈ C
∗, we get that:
σ(〈z, z′ − β〉) = 〈z − α′, z′〉 or 〈z, z′ − β′〉, (7)
where α′, β′ ∈ C∗.
We now use this information to prove that the action of Aut(U+
q(so5)) on the centre of
U+q
(so5) is trivial. More precisely, we are now in position to prove the following result.
Proposition 2.6. Let σ ∈ Aut(U+q
(so5)). There exist λ, λ′ ∈ C∗ such that
σ(z) = λz and σ(z′) = λ′z′.
Proof. We only prove the result for z. First, using the fact that U+q
(so5) is noetherian, it is
easy to show that, for any family βii∈N of pairwise distinct nonzero complex numbers, we
have:
〈z〉 =⋂
i∈N
P0,βiand 〈z′〉 =
⋂
i∈N
Pβi,0,
where Pα,β := 〈z − α, z′ − β〉. Indeed, if the inclusion
〈z〉 ⊆ I :=⋂
i∈N
P0,βi
is not an equality, then any P0,βiis a minimal prime over I for height reasons. As the P0,βi
are
pairwise distinct, I is a two-sided ideal of U+
q(so5) with infinitely many prime ideals minimal
over it. This contradicts the noetherianity of U+q
(so5). Hence
〈z〉 =⋂
i∈N
P0,βiand 〈z′〉 =
⋂
i∈N
Pβi,0,
and so
σ (〈z〉) =⋂
i∈N
σ(P0,βi).
It follows from (7) that, for all i ∈ N, there exists (γi, δi) 6= (0, 0) with γi = 0 or δi = 0
such that
σ(P0,βi) = Pγi,δi
.
Naturally, we can choose the family βii∈N such that either γi = 0 for all i ∈ N, or δi = 0
for all i ∈ N. Moreover, observe that, as the βi are pairwise distinct, so are the γi or the δi.
139
Hence, either
σ (〈z〉) =⋂
i∈N
Pγi,0,
or
σ (〈z〉) =⋂
i∈N
P0,δi,
that is,
either 〈σ(z)〉 = σ (〈z〉) = 〈z′〉 or 〈σ(z)〉 = σ (〈z〉) = 〈z〉.
As z, σ(z) and z′ are all central, it follows from Lemma 2.1 that there exists λ ∈ C∗ such
that either σ(z) = λz or σ(z) = λz′.
To conclude, it just remains to show that the second case cannot happen. In order to do
this, we use a graded argument. Observe that, with respect to the N-graduation of U+
q(so5)
defined in Section 2.2, z and z′ are homogeneous of degree 3 and 4 respectively. Thus, if
σ(z) = λz′, then we would obtain a contradiction with the fact that every automorphism
of U+q
(so5) preserves the valuation, see Corollary 2.4. Hence σ(z) = λz, as desired. The
corresponding result for z′ can be proved in a similar way, so we omit it.
Andruskiewitsch and Dumas, [4, Proposition 3.3], have proved that the subgroup of those
automorphisms of U+q
(so5) that stabilize 〈z〉 is isomorphic to (C∗)2. Thus, as we have just
shown that every automorphism of U+
q(so5) fixes 〈z〉, we get that Aut(U+
q(so5)) itself is
isomorphic to (C∗)2. This is the route that we have followed in [16] in order to prove the
Andruskiewitsch-Dumas Conjecture in the case where g = so5. Recently, with Samuel Lopes,
we proved this Conjecture in the case where g = sl4 using different methods and in particular
graded arguments. We are now using (similar) graded arguments to prove that every auto-
morphism of U+q
(so5) is a torus automorphism (witout using results of Andruskiewitsch and
Dumas).
In the proof, we will need the following relation that is easily obtained by straightforward
computations.
Lemma 2.7. (q2 − 1)E3E′3
= (q4 − 1)zE2 + q2z′.
Proposition 2.8. Let σ be an automorphism of U+q
(so5). Then there exist a1, b2 ∈ C∗ such
that
σ(E1) = a1E1 and σ(E2) = b2E2.
Proof. For all i ∈ 1, . . . , 4, we set di := deg(σ(Ei)). We also set d′3
:= deg(σ(E′3)). It follows
from Corollary 2.4 that d1, d2 ≥ 1, d3, d′3≥ 2 and d4 ≥ 3. First we prove that d1 = d2 = 1.
Assume first that d1 + d3 > 3. As z = (1 − q2)E1E3 + q2(q + q−1)E4 and σ(z) = λz with
λ ∈ C∗ by Proposition 2.6, we get:
λz = (1 − q2)σ(E1)σ(E3) + q2(q + q−1)σ(E4). (8)
140
Recall that deg(uv) = deg(u) + deg(v) for u, v 6= 0, as U+q
(g) is a domain. Thus, as deg(z) =
3 < deg(σ(E1)σ(E3)) = d1+d3, we deduce from (8) that d1+d3 = d4. As z′ = −(q2−q−2)(q+
q−1)E4E2 + q2(q2 − 1)E2
3and deg(z′) = 4 < d1 + d3 + d2 = d4 + d2 = deg(σ(E4)σ(E2)), we
get in a similar manner that d2 + d4 = 2d3. Thus d1 + d2 = d3. As d1 + d3 > 3, this forces
d3 > 2 and so d3 + d′3
> 4. Thus we deduce from Lemma 2.7 that d3 + d′3
= 3 + d2. Hence
d1 + d′3
= 3. As d1 ≥ 1 and d′3≥ 2, this implies d1 = 1 and d′
3= 2.
Thus we have just proved that d1 = deg(σ(E1)) = 1 and either d3 = 2 or d′3
= 2. To prove
that d2 = 1, we distinguish between these two cases.
If d3 = 2, then as previously we deduce from the relation z′ = −(q2−q−2)(q+q−1)E4E2 +
q2(q2 − 1)E2
3that d2 + d4 = 4, so that d2 = 1, as desired.
If d′3
= 2, then one can use the definition of E′3
and the previous expression of z′ in order to
prove that z′ = q−2(q2 − 1)E′2
3+ E2u, where u is a nonzero homogeneous element of U+
q(so5)
of degree 3. (u is nonzero since 〈z′〉 is a completely prime ideal and E′3
/∈ 〈z′〉 for degree
reasons.) As d′3
= 2 and deg(σ(z′)) = 4, we get as previously that d2 = 1.
To summarise, we have just proved that deg(σ(E1)) = 1 = deg(σ(E2)), so that σ(E1) =
a1E1 + a2E2 and σ(E2) = b1E1 + b2E2, where (a1, a2), (b1, b2) ∈ C2 \ (0, 0). To conclude
that a2 = b1 = 0, one can for instance use the fact that σ(E1) and σ(E2) must satisfy the
quantum Serre relations.
We have just confirmed the Andruskiewitsch-Dumas Conjecture in the case where g = so5.
Theorem 2.9. Every automorphism of U+q
(so5) is a torus automorphism, so that
Aut(U+
q(so5)) ≃ (C∗)2.
2.5 Beyond these two cases
To finish this overview paper, let us mention that recently the Andruskiewitsch-Dumas Con-
jecture was confirmed by Samuel Lopes and the author, [18], in the case where g = sl4. The
crucial step of the proof is to prove that, up to an element of G, every normal element of
U+q
(sl4) is fixed by every automorphism. This step was dealt with by first computing the Lie
algebra of derivations of U+
q(sl4), and this already requires a lot of computations!
Acknowledgements
I thank Jacques Alev, Francois Dumas, Tom Lenagan and Samuel Lopes for all the interesting
conversations that we have shared on the topics of this paper. I also like to thank the organ-
isers of the Workshop ”From Lie Algebras to Quantum Groups” (and all the participants) for
this wonderful meeting. Finally, I would like to express my gratitude for the hospitality re-
ceived during my subsequent visit to the University of Porto, especially from Paula Carvalho
Lomp, Christian Lomp and Samuel Lopes.
141
References
[1] J. Alev, Un automorphisme non modere de U(g3), Comm. Algebra 14 (8), 1365-
1378 (1986).
[2] J. Alev and F. Dumas, Rigidite des plongements des quotients primitifs mini-
maux de Uq(sl(2)) dans l’algebre quantique de Weyl-Hayashi, Nagoya Math. J. 143
(1996), 119-146.
[3] J. Alev and F. Dumas, Sur le corps des fractions de certaines algebres quantiques,
J. Algebra 170 (1994), 229-265.
[4] N. Andruskiewitsch and F. Dumas, On the automorphisms of U+q
(g),
ArXiv:math.QA/0301066, to appear.
[5] K.A. Brown and K.R. Goodearl, Lectures on algebraic quantum groups. Advanced
Courses in Mathematics-CRM Barcelona. Birkhauser Verlag, Basel, 2002.
[6] K.A. Brown and K.R. Goodearl, Prime spectra of quantum semisimple groups,
Trans. Amer. Math. Soc. 348 (1996), no. 6, 2465-2502.
[7] P. Caldero, Sur le centre de Uq(n+), Beitrage Algebra Geom. 35 (1994), no. 1,
13-24.
[8] P. Caldero, Etude des q-commutations dans l’algebre Uq(n+), J. Algebra 178 (1995),
444-457.
[9] O. Fleury, Automorphismes de Uq(b+), Beitrage Algebra Geom. 38 (1994), 13-24.
[10] K.R. Goodearl and E.S. Letzter, Prime factor algebras of the coordinate ring of
quantum matrices, Proc. Amer. Math. Soc. 121 (1994), no. 4, 1017-1025.
[11] K.R. Goodearl and E.S. Letzter, The Dixmier-Moeglin equivalence in quantum
coordinate rings and quantized Weyl algebras, Trans. Amer. Math. Soc. 352 (2000),
no. 3, 1381-1403.
[12] M. Gorelik, The prime and the primitive spectra of a quantum Bruhat cell translate,
J. Algebra 227 (2000), no. 1, 211-253.
[13] J.C. Jantzen, Lectures on Quantum Groups, in: Grad. Stud. Math., Vol. 6, Amer.
Math. Society, Providence, RI, 1996.
[14] A. Joseph, A wild automorphism of U(sl2), Math. Proc. Camb. Phil. Soc. (1976),
80, 61-65.
142
[15] A. Joseph, Quantum groups and their primitive ideals. Springer-Verlag, 29, Ergeb-
nisse der Mathematik und ihrer Grenzgebiete, 1995.
[16] S. Launois, Primitive ideals and automorphism group of U+
q(B2), J. Algebra Appl.
6 (2007), no. 1, 21-47.
[17] S. Launois and T.H. Lenagan, Primitive ideals and automorphisms of quan-
tum matrices, to appear in Algebras and Representation Theory, posted at
arXiv:math.RA/0511409.
[18] S. Launois and S.A. Lopes, Automorphisms and derivations of Uq(sl+
4), to appear
in Journal of Pure and Applied Algebra, posted at arXiv:math.QA/0606134.
[19] G. Lusztig, Introduction to quantum groups, Progress in Mathematics, 110,
Birkhauser Boston, Inc., Boston, MA, 1993.
[20] M.P. Malliavin, L’algebre d’Heisenberg quantique, Bull. Sci. Math. 118 (1994), 511-
537.
[21] C.M. Ringel, PBW-bases of quantum groups, J. Reine Angew. Math. 470 (1996),
51-88.
[22] I.P. Shestakov and U.U. Umirbaev, The tame and the wild automorphisms of poly-
nomial rings in three variables, J. Amer. Math. Soc. 17 (2004), no. 1, 197-227
143
144
An Artinian theory for Lie algebras
Antonio Fernandez Lopez∗ Esther Garcıa† Miguel Gomez Lozano‡
Abstract
We summarize in this contribution the main results of our paper [5]. Only those
definitions which are necessary to understand the statements are provided, but no proofs,
which can be found in [5].
1 Lie algebras and Jordan pairs
1. Throughout this paper, and at least otherwise specified, we will be dealing with Lie al-
gebras L [7], [11], and Jordan pairs V = (V +, V −) [8], over a ring of scalars Φ containing1
30.
2. Let V = (V +, V −) be a Jordan pair. An element x ∈ V σ, σ = ±, is called an absolute
zero divisor if Qx = 0, and V is said to be nondegenerate if it has no nonzero absolute
zero divisors. Similarly, given a Lie algebra L, x ∈ L is an absolute zero divisor if
ad2
x= 0, L is nondegenerate if it has no nonzero absolute zero divisors.
3. Given a Jordan pair V = (V +, V −), an inner ideal of V is any Φ-submodule B of
V σ such that B,V −σ, B ⊂ B. Similarly, an inner ideal of a Lie algebra L is a Φ-
submodule B of L such that [B, [B,L]] ⊂ B. An abelian inner ideal is an inner ideal B
which is also an abelian subalgebra, i.e., [B,B] = 0.
4. (a) Recall that the socle of a nondegenerate Jordan pair V is SocV = (Soc V +,SocV −)
where SocV σ is the sum of all minimal inner ideals of V contained in V σ [9]. The
socle of a nondegenerate Lie algebra L is SocL, defined as the sum of all minimal
inner ideals of L [3].
∗Departamento de Algebra, Geometrıa y Topologıa, Universidad de Malaga, 29071 Malaga, Spain. Email:
[email protected]. Partially supported by the MEC and Fondos FEDER, MTM2004-03845, and by
Junta de Andalucıa FQM264.†Departamento de Matematica Aplicada, Universidad Rey Juan Carlos, 28933 Mostoles. Madrid, Spain.
Email: [email protected]. Partially supported by the MEC and Fondos FEDER, MTM2004-06580-
C02-01, and by FICYT-IB05-017.‡Departamento de Algebra, Geometrıa y Topologıa, Universidad de Malaga, 29071 Malaga, Spain. Email:
[email protected]. Partially supported by the MEC and Fondos FEDER, MTM2004-03845, by Junta
de Andalucıa FQM264, and by FICYT-IB05-017.
145
(b) By [9, Theorem 2] (for the Jordan pair case) and [3, Theorem 2.5] (for the Lie
case), the socle of a nondegenerate Jordan pair or Lie algebra is the direct sum of
its simple ideals. Moreover, each simple component of Soc L is either inner simple
or contains an abelian minimal inner ideal [2, Theorem 1.12].
(c) A Lie algebra L or Jordan pair V is said to be Artinian if it satisfies the descending
chain condition on all inner ideals.
(d) By definition, a properly ascending chain 0 ⊂ M1 ⊂ M2 ⊂ · · · ⊂ Mn of inner ideals
of a Lie algebra L has length n. The length of an inner ideal M is the supremum
of the lengths of chains of inner ideals of L contained in M .
2 Complemented Lie algebras
The module-theoretic characterization of semiprime Artinian rings (R is unital and completely
reducible as a left R-module) cannot be translated to Jordan systems by merely replacing
left ideals by inner ideals: if we take, for instance, the Jordan algebra M2(F )(+) of 2 × 2-
matrices over a field F , any nontrivial inner ideal of M2(F )(+) has dimension 1, so it cannot be
complemented as a F -subspace by any other inner ideal. Nevertheless, O. Loos and E. Neher
succeeded in getting the appropriate characterization by introducing the notion of kernel of
an inner ideal [10]:
A Jordan pair V = (V +, V −) (over an arbitrary ring of scalars) is a direct sum of simple
Artinian nondegenerate Jordan pairs if and only if it is complemented in the following sense:
for any inner ideal B of V σ there exists an inner ideal C of V −σ such that each of them
is complemented as a submodule by the kernel of the other. In particular, a simple Jordan
pair is complemented if and only if is nondegenerate and Artinian. A similar characterization
works for Lie algebras.
1.
Let M be an inner ideal of a Lie algebra L. The kernel of M
Ker M = x ∈ L : [M, [M,x]] = 0
is a Φ-submodule of L. An inner complement of M is an inner ideal N of L such that
L = M ⊕ KerN = N ⊕ KerM.
A Lie algebra L will be called complemented if any inner ideal of L has an inner comple-
ment, and abelian complemented if any abelian inner ideal has an inner complement which is
abelian. Our main result, which can be regarded as a Lie analogue of the module-theoretic
characterization of semiprime Artinian rings, proves.
Theorem 2. For a Lie algebra L, the following notions are equivalent:
146
(i) L is complemented.
(ii) L is a direct sum of ideals each of which is a simple nondegenerate Artinian Lie algebra.
Moreover,
(iii) Complemented Lie algebras are abelian complemented.
A key tool used in the proof of (i) ⇒ (ii) is the notion of subquotient of a Lie algebra with
respect to an abelian inner ideal:
3.
For any abelian inner ideal M of L, the pair of Φ-modules
V = (M,L/KerL M)
with the triple products given by
m,a, n := [[m,a], n] for every m,n ∈ M and a ∈ L
a,m, b := [[a,m], b] for every m ∈ M and a, b ∈ L,
where x denotes the coset of x relative to the submodule KerL M , is a Jordan pair called the
subquotient of L with respect to M .
Subquotients inherit, on the one hand, regularity conditions from the Lie algebra, and,
on the other hand, keep the inner ideal structure of L within them. This fact turns out to
be crucial for using results of Jordan theory. For instance, it is used to prove that any prime
abelian complemented Lie algebra satisfies the ascending and descending chain conditions on
abelian inner ideals. Moreover, as proved in [6], an abelian inner ideal B of finite length in
a nondegenerate Lie algebra L is not just complemented by abelian inner ideals, but there
exists a short grading L = L−n ⊕ · · · ⊕ L0 ⊕ · · · ⊕ Ln such that B = Ln, and hence B is
complemented by L−n.
It must be noted that, while any nondegenerate Artinian Jordan pair is a direct sum
of finitely many simple nondegenerate Artinian Jordan pairs, a nondegenerate Artinian Lie
algebra does only have essential socle [3, Corollary 2.6]. In fact, there exist strongly prime
finite dimensional Lie algebras (over a field of characteristic p > 5) with nontrivial ideals
[12, p. 152]. Therefore, unlike the Jordan case, nondegenerate Artinian Lie algebras are not
necessarily complemented: they are only abelian complemented.
3 Simple nondegenerate Artinian Lie algebras
Let L be a simple nondegenerate Lie algebra containing an abelian minimal inner ideal. Then
L has a 5-grading. Hence, by [13, Theorem 1], L is one of the following: (i) a simple Lie algebra
147
of type G2, F4, E6, E7 or E8, (ii) L = R′= [R,R]/Z(R)∩[R,R], where R is a simple associative
algebra such that [R,R] is not contained in Z(R), or (iii) L = K′= [K,K]/Z(R) ∩ [K,K],
where K = Skew(R, ∗) and R is a simple associative algebra, ∗ is an involution of R, and either
Z(R) = 0 or the dimension of R over Z(R) is greater than 16. (Actually, the list of simple
Lie algebras with gradings given in [13, Theorem 1], contains two additional algebras: the
Tits-Kantor-Koecher algebra of a nondegenerate symmetric bilinear form and D4. However,
because of we are not interested in describing the gradings, both algebras can be included
in case (iii): K′
= K ′ = K = Skew(R, ∗), where R is a simple algebra with orthogonal
involution.)
By using the inner ideal structure of Lie algebras of traceless operators of finite rank which
are continuous with respect to an infinite dimensional pair of dual vector spaces over a division
algebra, and that of the Lie algebras of finite rank skew operators on an infinite dimensional
self dual vector space (extending the work of G. Benkart [1] for the finite dimensional case,
and a previous one of the authors [4] for finitary Lie algebras), we can refine the above list in
the case of a simple nondegenerate Artinian Lie algebra.
4.
A Lie algebra L will be called a division Lie algebra if it is nonzero, nondegenerate and
has no nontrivial inner ideals. Two examples are given below:
1. Let ∆ be a division associative algebra such that [[∆,∆],∆] 6= 0. Then [∆,∆]/[∆,∆]∩
Z(∆) is a division Lie algebra, [1, Corollary 3.15].
2. Let R be a simple associative algebra with involution ∗ and nonzero socle. Suppose that
Z(R) = 0 or the dimension of R over Z(R) is greater than 16, and set K := Skew(R, ∗).
Then L = [K,K]/[K,K] ∩ Z(R) is a division Lie algebra if and only if (R, ∗) has no
nonzero isotropic right ideals, i.e., those right ideals I such that I∗I = 0. This is a direct
consequence of the inner ideal structure of L: [1, Theorem 5.5] when R is Artinian, and
[4] when R is not Artinian.
Theorem 5. Let L be a simple Lie algebra over a field F of characteristic 0 or greater than
7. Then L is Artinian and nondegenerate if and only if it is one of the following:
1. A division Lie algebra.
2. A simple Lie algebra of type G2, F4, E6, E7 or E8 containing abelian minimal inner
ideals.
3. [R,R]/[R,R]∩Z(R), where R is a simple Artinian (but not division) associative algebra.
4. [K,K]/[K,K]∩Z(R), where K = Skew(R, ∗) and R is a simple associative algebra with
involution ∗ which coincides with its socle, such that Z(R) = 0 or the dimension of
148
R over Z(R) is greater than 16, and (R, ∗) satisfies the descending chain condition on
isotropic right ideals.
Summarizing, we can say that, as conjectured by G. Benkart in the introduction of [2],
inner ideals in Lie algebras play a role analogous to Jordan inner ideals in the development
of an Artinian theory for Lie algebras.
References
[1] G. Benkart, The Lie inner ideal structure of associative rings, J. Algebra, 43, (1976),
561-584.
[2] G. Benkart, On inner ideals and ad-nilpotent elements of Lie algebras, Trans. Amer.
Math. Soc., 232, (1977), 61-81.
[3] C. Draper, A. Fernandez Lopez, E. Garcıa and M. Gomez Lozano,
The socle of a nondegenerate Lie algebra, Preprint, available as paper [185]
of the Jordan Theory Preprint Archives (http://www.uibk.ac.at/ mathematik
/loos/jordan/index.html).
[4] A. Fernandez Lopez, E. Garcıa and M. Gomez Lozano, Inner ideals of simple finitary
Lie algebras, J. Lie Theory, 16, (2006), 97-114.
[5] A. Fernandez Lopez, E. Garcıa and M. Gomez Lozano, An Artinian theory for Lie
algebras, Preprint, available as paper [217] of the Jordan Theory Preprint Archives (http:
//www.uibk.ac.at/mathematik/loos/jordan/index.html).
[6] A. Fernandez Lopez, E. Garcıa, M. Gomez Lozano and E. Neher, A construction of
gradings of Lie algebras, Preprint, available as paper [226] of the Jordan Theory Preprint
Archives (http://www.uibk.ac.at/mathematik/loos/jordan/index.html).
[7] N. Jacobson, Lie Algebras, Interscience Publishers, New York, (1962).
[8] O. Loos, Jordan Pairs, Lectures Notes in Mathematics, Vol. 460, Springer-Verlag, New
York, (1975).
[9] O. Loos, On the socle of a Jordan pair, Collect. Math., 40, no. 2, (1989), 109–125.
[10] O. Loos and E. Neher, Complementation of inner ideals in Jordan pairs, J. Algebra,
166 (1994), 255-295.
[11] E. Neher, Lie algebras graded by 3-graded root systems and Jordan pairs covered by grids,
Amer. J. Math., 118, (1996), 439-491.
149
[12] A. A. Premet, Lie algebras without strong degeneration, Math USSR Sbornik, 57, (1987),
151-164.
[13] E. I. Zelmanov, Lie algebras with a finite grading, Math USSR Sbornik, 52, (1985),
347-385.
150
Quantum groupoids with projection
J.N. Alonso Alvarez∗ J.M. Fernandez Vilaboa† R. Gonzalez Rodrıguez‡
Abstract
In this survey we explain in detail how Radford’s ideas and results about Hopf algebras
with projection can be generalized to quantum groupoids in a strict symmetric monoidal
category with split idempotents.
Introduction
Let H be a Hopf algebra over a field K and let A be a K-algebra. A well-known result of
Radford [23] gives equivalent conditions for an object A ⊗H equipped with smash product
algebra and coalgebra to be a Hopf algebra and characterizes such objects via bialgebra
projections. Majid in [16] interpreted this result in the modern context of Yetter-Drinfeld
modules and stated that there is a correspondence between Hopf algebras in this category,
denoted by H
HYD, and Hopf algebras B with morphisms of Hopf algebras f : H → B,
g : B → H such that g f = idH . Later, Bespalov proved the same result for braided
categories with split idempotents in [5]. The key point in Radford-Majid-Bespalov’s theorem
is to define an object BH , called the algebra of coinvariants, as the equalizer of (B ⊗ g) δB
and B ⊗ ηH . This object is a Hopf algebra in the category H
HYD and there exists a Hopf
algebra isomorphism between B and BH ⊲⊳ H (the smash (co)product of BH and H). It is
important to point out that in the construction of BH ⊲⊳ H they use that BH is the image of
the idempotent morphism qB
H= µB (B ⊗ (f λH g)) δB .
In [11], Bulacu and Nauwelaerts generalize Radford’s theorem about Hopf algebras with
projection to the quasi-Hopf algebra setting. Namely, if H and B are quasi-Hopf algebras
with bijective antipode and with morphisms of quasi-Hopf algebras f : H → B, g : B → H
such that g f = idH , then they define a subalgebra Bi (the generalization of BH to this
setting) and with some additional structures Bi becomes, a Hopf algebra in the category
of left-left Yetter-Drinfeld modules H
HYD defined by Majid in [17]. Moreover, as the main
∗Departamento de Matematicas, Universidad de Vigo, Campus Universitario Lagoas-Marcosende, E-36280
Vigo, Spain. E-mail: [email protected]†Departamento de Alxebra, Universidad de Santiago de Compostela. E-15771 Santiago de Compostela,
Spain. E-mail: [email protected]‡Departamento de Matematica Aplicada II, Universidad de Vigo, Campus Universitario Lagoas-
Marcosende, E-36310 Vigo, Spain. E-mail: [email protected]
151
result in [11], Bulacu and Nauwelaerts state that Bi × H is isomorphic to B as quasi-Hopf
algebras where the algebra structure of Bi ×H is the smash product defined in [10] and the
quasi-coalgebra structure is the one introduced in [11].
The basic motivation of this survey is to explain in detail how the above ideas and results
can be generalized to quantum groupoids in a strict symmetric monoidal category with split
idempotents. Quantum groupoids or weak Hopf algebras have been introduced by Bohm, Nill
and Szlachanyi [7] as a new generalization of Hopf algebras and groupoid algebras. Roughly
speaking, a weak Hopf algebra H in a symmetric monoidal category is an object that has
both algebra and coalgebra structures with some relations between them and that possesses
an antipode λH which does not necessarily verify λH ∧ idH = idH ∧ λH = εH ⊗ ηH where εH ,
ηH are the counity and unity morphisms respectively and ∧ denotes the usual convolution
product. The main differences with other Hopf algebraic constructions, such as quasi-Hopf
algebras and rational Hopf algebras, are the following: weak Hopf algebras are coassociative
but the coproduct is not required to preserve the unity ηH or, equivalently, the counity is
not an algebra morphism. Some motivations to study weak Hopf algebras come from their
connection with the theory of algebra extensions, the important applications in the study of
dynamical twists of Hopf algebras and their link with quantum field theories and operator
algebras (see [20]).
The survey is organized as follows.
In Section 1 we give basis definitions and examples of quantum groupoids without finite-
ness conditions. Also we introduce the category of left-left Yetter-Drinfeld modules defined
by Bohm for a quantum groupoid with invertible antipode. As in the case of Hopf algebras
this category is braided monoidal but in this case is not strict.
The exposition of the theory of crossed products associated to projections of quantum
groupoids in Section 2 follows [2] and is the good generalization of the classical theory devel-
oped by Blattner, Cohen and Montgomery in [6]. The main theorem in this section generalizes
a well know result, due to Blattner, Cohen and Montgomery, which shows that if Bπ→ H → 0
is an exact sequence of Hopf algebras with coalgebra splitting then B ≈ A♯σH, where A is the
left Hopf kernel of π and σ is a suitable cocycle (see Theorem (4.14) of [6]). In this section we
show that if g : B → H is a morphism of quantum groupoids and there exists a morphism of
coalgebras f : H → B such that gf = idH and f ηH = ηB , using the idempotent morphism
qB
H= µB (B ⊗ (λB f g)) δB : B → B it is possible to construct an equalizer diagram
and an algebra BH , i.e, the algebra of coinvariants or the Hopf kernel of g, and morphisms
ϕBH: H ⊗ BH → BH (the weak measuring), σBH
: H ⊗ H → BH (the weak cocycle) such
that there exists an idempotent endomorphism of BH ⊗H which image, denoted by BH ×H,
is isomorphic with B as algebras being the algebra structure (crossed product algebra)
ηBH×H = rB (ηBH⊗ ηH),
µBH×H = rB (µBH⊗H) (µBH
⊗ σBH⊗ µH) (BH ⊗ ϕBH
⊗ δH⊗H)
152
(BH ⊗H ⊗ cH,BH⊗H) (BH ⊗ δH ⊗BH ⊗H) (sB ⊗ sB),
where sB is the inclusion of BH × H in BH ⊗ H and rB the projection of BH ⊗ H on
BH ×H. Of course, when H, B are Hopf algebras we recover the result of Blattner, Cohen
and Montgomery. For this reason, we denote the algebra BH×H by BH♯σBHH. If moreover f
is an algebra morphism, the cocycle is trivial in a weak sense and then we obtain that µBH×H
is the weak version of the smash product used by Radford in the Hopf algebra setting. Also,
we prove the dual results using similar arguments but passing to the opposite category, for a
morphism of quantum groupoids h : H → B and an algebra morphism t : B → H such that
t h = idH and εH t = εB .
Finally, in Section 3, linking the information of section 2 with the results of [1], [2], [3] and
[4], we obtain our version of Radford’s Theorem for quantum groupoids with projection. In
this section we prove that the algebra of coinvariants BH associated to a quantum groupoid
projection (i.e. a pair of morphisms of quantum groupoids f : H → B, g : B → H such that
g f = idH) can be obtained as an equalizer or, by duality, as a coequalizer (in this case the
classical theory developed in Section 2 and the dual one provide the same object BH with dual
algebraic structures, algebra-coalgebra, module-comodule, etc...). Therefore, it is possible to
find an algebra coalgebra structure for BH and morphisms ϕBH= pB
H µB (f ⊗ iB
H) :
H ⊗BH → BH and BH= (g ⊗ pB
H) δB iB
H: BH → H ⊗BH such that (BH , ϕBH
) is a left
H-module and (BH , BH) is a left H-comodule. We show that BH is a Hopf algebra in the
category of left-left Yetter-Drinfeld modules H
HYD and, using the the the weak smash product
and the weak smash coproduct of BH and H we give a good weak Hopf algebra interpretation
of the theorems proved by Radford [23] and Majid [16] in the Hopf algebra setting, obtaining
an isomorphism of quantum groupoids between BH ×H and B.
1 Quantum groupoids in monoidal categories
In this section we give definitions and discuss basic properties of quantum groupoids in
monoidal categories.
Let C be a category. We denote the class of objects of C by |C| and for each object X ∈ |C|,
the identity morphism by idX : X → X.
A monoidal category (C,⊗,K, a, l, r) is a category C which is equipped with a tensor
product ⊗ : C × C → C, with an object K, called the unit of the monoidal category, with a
natural isomorphism a : ⊗(id×⊗) → ⊗(⊗× id), called the associativity constrain, and with
natural isomorphisms l : ⊗(K × id) → id, r : ⊗(id×K) → id, called left unit constraint and
right unit constraint respectively, such that the Pentagon Axiom
(aU,V,W ⊗ idX) aU,V ⊗W,X (idU ⊗ aV,W,X) = aU⊗V,W,X aU,V,W⊗X
and the Triangle Axiom
idV ⊗ lW = (rV ⊗ idW ) aV,K,W
153
are satisfied.
The monoidal category is said to be strict if the associativity and the unit constraints a,
l, r are all identities of the category.
Let Ψ : C×C → C×C be the flip functor defined by Ψ(V,W ) = (W,V ) on any pair of objects
of C. A commutativity constrain is a natural isomorphism c : ⊗ → ⊗Ψ. If (C,⊗,K, a, l, r)
is a monoidal category, a braiding in C is a commutativity constraint satisfying the Hexagon
Axiom
aW,U,V cU⊗V,W aU,V,W = (cU,W ⊗ idV ) aU,W,V (idU ⊗ cV,W ),
a−1
V,W,U cU,V ⊗W a−1
U,V,W= (idV ⊗ cU,W ) a−1
V,U,W (cU,V ⊗ idW ).
A braided monoidal category is a monoidal category with a braiding c. These categories
generalizes the classical notion of symmetric monoidal category introduced earlier by category
theorists. A braided monoidal category is symmetric if the braiding satisfies cW,V cV,W =
idV ⊗W for all V,W ∈ |C|.
From now on we assume that C is strict symmetric and admits split idempotents, i.e.,
for every morphism ∇Y : Y → Y such that ∇Y = ∇Y ∇Y there exist an object Z and
morphisms iY : Z → Y and pY : Y → Z such that ∇Y = iY pY and pY iY = idZ . There
is not loss of generality in assuming the strict character for C because it is well know that
given a monoidal category we can construct a strict monoidal category Cst which is tensor
equivalent to C (see [15] for the details). For simplicity of notation, given objects M , N , P
in C and a morphism f : M → N , we write P ⊗ f for idP ⊗ f and f ⊗ P for f ⊗ idP .
Definition 1.1. An algebra in C is a triple A = (A, ηA, µA) where A is an object in C and
ηA : K → A (unit), µA : A⊗A→ A (product) are morphisms in C such that µA (A⊗ ηA) =
idA = µA (ηA ⊗A), µA (A⊗µA) = µA (µA ⊗A). Given two algebras A = (A, ηA, µA) and
B = (B, ηB , µB), f : A → B is an algebra morphism if µB (f ⊗ f) = f µA, f ηA = ηB .
Also, if A, B are algebras in C, the object A ⊗ B is an algebra in C where ηA⊗B = ηA ⊗ ηB
and µA⊗B = (µA ⊗ µB) (A⊗ cB,A ⊗B).
A coalgebra in C is a triple D = (D, εD, δD) where D is an object in C and εD : D → K
(counit), δD : D → D⊗D (coproduct) are morphisms in C such that (εD ⊗D) δD = idD =
(D ⊗ εD) δD, (δD ⊗ D) δD = (D ⊗ δD) δD. If D = (D, εD, δD) and E = (E, εE , δE)
are coalgebras, f : D → E is a coalgebra morphism if (f ⊗ f) δD = δE f , εE f = εD.
When D, E are coalgebras in C, D ⊗ E is a coalgebra in C where εD⊗E = εD ⊗ εE and
δD⊗E = (D ⊗ cD,E ⊗ E) (δD ⊗ δE).
If A is an algebra, B is a coalgebra and α : B → A, β : B → A are morphisms, we define
the convolution product by α ∧ β = µA (α⊗ β) δB .
By quantum groupoids or weak Hopf algebras we understand the objects introduced in
[7], as a generalization of ordinary Hopf algebras. Here, for the convenience of the reader, we
recall the definition of these objects and some relevant results from [7] without proof, thus
making our exposition self-contained.
154
Definition 1.2. A quantum groupoidH is an object in C with an algebra structure (H, ηH , µH)
and a coalgebra structure (H, εH , δH) such that the following axioms hold:
(a1) δH µH = (µH ⊗ µH) δH⊗H ,
(a2) εH µH (µH ⊗H) = (εH ⊗ εH) (µH ⊗ µH) (H ⊗ δH ⊗H)
= (εH ⊗ εH) (µH ⊗ µH) (H ⊗ (cH,H δH) ⊗H),
(a3) (δH ⊗H) δH ηH = (H ⊗ µH ⊗H) (δH ⊗ δH) (ηH ⊗ ηH)
= (H ⊗ (µH cH,H) ⊗H) (δH ⊗ δH) (ηH ⊗ ηH).
(a4) There exists a morphism λH : H → H in C (called the antipode of H) verifiying:
(a4-1) idH ∧ λH = ((εH µH) ⊗H) (H ⊗ cH,H) ((δH ηH) ⊗H),
(a4-2) λH ∧ idH = (H ⊗ (εH µH)) (cH,H ⊗H) (H ⊗ (δH ηH)),
(a4-3) λH ∧ idH ∧ λH = λH .
Note that, in this definition, the conditions (a2), (a3) weaken the conditions of multiplica-
tivity of the counit, and comultiplicativity of the unit that we can find in the Hopf algebra
definition. On the other hand, axioms (a4-1), (a4-2) and (a4-3) weaken the properties of
the antipode in a Hopf algebra. Therefore, a quantum groupoid is a Hopf algebra if an only
if the morphism δH (comultiplication) is unit-preserving and if and only if the counit is a
homomorphism of algebras.
1.3. If H is a quantum groupoid in C, the antipode λH is unique, antimultiplicative, antico-
multiplicative and leaves the unit ηH and the counit εH invariant:
λH µH = µH (λH ⊗ λH) cH,H , δH λH = cH,H (λH ⊗ λH) δH ,
λH ηH = ηH , εH λH = εH .
If we define the morphisms ΠL
H(target morphism), ΠR
H(source morphism), Π
L
Hand Π
R
H
by
ΠL
H= ((εH µH) ⊗H) (H ⊗ cH,H) ((δH ηH) ⊗H),
ΠR
H= (H ⊗ (εH µH)) (cH,H ⊗H) (H ⊗ (δH ηH)),
ΠL
H= (H ⊗ (εH µH)) ((δH ηH) ⊗H),
ΠR
H= ((εH µH) ⊗H) (H ⊗ (δH ηH)).
155
it is straightforward to show that they are idempotent and ΠL
H, ΠR
Hsatisfy the equalities
ΠL
H= idH ∧ λH , ΠR
H= λH ∧ idH .
Moreover, we have that
ΠL
H Π
L
H= ΠL
H, ΠL
H Π
R
H= Π
R
H, ΠR
H Π
L
H= Π
L
H, ΠR
H Π
R
H= ΠR
H,
ΠL
H ΠL
H= Π
L
H, Π
L
H ΠR
H= ΠR
H, Π
R
H ΠL
H= ΠL
H, Π
R
H ΠR
H= Π
R
H.
Also it is easy to show the formulas
ΠL
H= Π
R
H λH = λH Π
L
H, ΠR
H= Π
L
H λH = λH Π
R
H,
ΠL
H λH = ΠL
H ΠR
H= λH ΠR
H, ΠR
H λH = ΠR
H ΠL
H= λH ΠL
H.
If λH is an isomorphism (for example, when H is finite), we have the equalities:
ΠL
H= µH (H ⊗ λ−1
H) cH,H δH , Π
R
H= µH (λ−1
H⊗H) cH,H δH .
If the antipode of H is an isomorphism, the opposite operator and the coopposite operator
produce quantum groupoids from quantum groupoids. In the first one the product µH is
replaced by the opposite product µHop = µH cH,H while in the second the coproduct δH is
replaced by δHcoop = cH,H δH . In both cases the antipode λH is replaced by λ−1
H.
A morphism between quantum groupoids H and B is a morphism f : H → B which is
both algebra and coalgebra morphism. If f : H → B is a weak Hopf algebra morphism, then
λB f = f λH (see Proposition 1.4 of [1]).
Examples 1.4. (i) As group algebras and their duals are the natural examples of Hopf
algebras, groupoid algebras and their duals provide examples of quantum groupoids. Recall
that a groupoid G is simply a category in which every morphism is an isomorphism. In this
example, we consider finite groupoids, i.e. groupoids with a finite number of objects. The
set of objects of G will be denoted by G0 and the set of morphisms by G1. The identity
morphism on x ∈ G0 will also be denoted by idx and for a morphism σ : x → y in G1, we
write s(σ) and t(σ), respectively for the source and the target of σ.
Let G be a groupoid, and R a commutative ring. The groupoid algebra is the direct
product
RG =⊕
σ∈G1
Rσ
with the product of two morphisms being equal to their composition if the latter is defined
and 0 in otherwise, i.e. στ = σ τ if s(σ) = t(τ) and στ = 0 if s(σ) 6= t(τ). The unit element
is 1RG =∑
x∈G0idx. The algebra RG is a cocommutative quantum groupoid, with coproduct
δRG, counit εRG and antipode λRG given by the formulas:
δRG(σ) = σ ⊗ σ, εRG(σ) = 1, λRG(σ) = σ−1.
156
For the quantum groupoid RG the morphisms target and source are respectively,
ΠL
RG(σ) = idt(σ), ΠR
RG(σ) = ids(σ)
and λRG λRG = idRG, i.e. the antipode is involutive.
If G1 is finite, then RG is free of a finite rank as a R-module, hence GR = (RG)∗ =
HomR(RG,R) is a commutative quantum groupoid with involutory antipode. As R-module
GR =⊕
σ∈G1
Rfσ
with 〈fσ, τ〉 = δσ,τ . The algebra structure is given by the formulas fσfτ = δσ,τfσ and 1GR =∑
σ∈G1fσ. The coalgebra structure is
δGR(fσ) =∑
τρ=σ
fτ ⊗ fρ =∑
ρ∈G1
fσρ−1 ⊗ fρ, εGR(fσ) = δσ,idt(σ).
The antipode is given by λGR(fσ) = fσ−1 .
(ii) It is known that any group action on a set gives rise to a groupoid (see [24]). In [20]
Nikshych and Vainerman extend this construction associating a quantum groupoid with any
action of a Hopf algebra on a separable algebra.
(iii) It was shown in [19] that any inclusion of type Π1 factors with finite index and depth
give rise to a quantum groupoid describing the symmetry of this inclusion. In [20] can be
found an example of this construction applied to the case of Temperley-Lieb algebras (see
[13]).
(iv) In [22] Nill proved that Hayashi’s face algebras [14] are examples of quantum groupoids
whose counital subalgebras, i.e., the images of ΠL
Hand ΠR
H, are commutative. Also, in [22]
we can find that Yamanouchi’s generalized Kac algebras (see [25]) are exactly C∗-quantum
groupoids with involutive antipode.
1.5. Let H be a quantum groupoid. We say that (M,ϕM ) is a left H-module if M is an
object in C and ϕM : H ⊗M → M is a morphism in C satisfying ϕM (ηH ⊗M) = idM ,
ϕM (H⊗ϕM ) = ϕM (µH⊗M). Given two leftH-modules (M,ϕM ) and (N,ϕN ), f : M → N
is a morphism of left H-modules if ϕN (H ⊗ f) = f ϕM . We denote the category of right
H-modules by HC. In an analogous way we define the category of right H-modules and we
denote it by CH .
If (M,ϕM ) and (N,ϕN ) are left H-modules we denote by ϕM⊗N the morphism ϕM⊗N :
H ⊗M ⊗N →M ⊗N defined by
ϕM⊗N = (ϕM ⊗ ϕN ) (H ⊗ cH,M ⊗N) (δH ⊗M ⊗N).
We say that (M,M ) is a left H-comodule if M is an object in C and M : M → H ⊗M
is a morphism in C satisfying (εH ⊗M) M = idM , (H ⊗ M ) M = (δH ⊗M) M . Given
157
two left H-comodules (M,M ) and (N, N ), f : M → N is a morphism of left H-comodules
if N f = (H ⊗ f) M . We denote the category of left H-comodules by HC. Analogously,
CH denotes the category of right H-comodules.
For two leftH-comodules (M,M ) and (N, N ), we denote by M⊗N the morphism M⊗N :
M ⊗N → H ⊗M ⊗N defined by
M⊗N = (µH ⊗M ⊗N) (H ⊗ cM,H ⊗N) (M ⊗ N ).
Let (M,ϕM ), (N,ϕN ) be left H-modules. Then the morphism
∇M⊗N = ϕM⊗N (ηH ⊗M ⊗N) : M ⊗N →M ⊗N
is idempotent. In this setting we denote by M × N the image of ∇M⊗N and by pM,N :
M ⊗N →M ×N , iM,N : M ×N →M ⊗N the morphisms such that iM,N pM,N = ∇M⊗N
and pM,N iM,N = idM×N . Using the definition of × we obtain that the object M ×N is a
left H-module with action ϕM×N = pM,N ϕM⊗N (H ⊗ iM,N ) : H ⊗ (M × N) → M × N
(see [20]). Note that, if f : M →M ′ and g : N → N ′ are morphisms of left H-modules then
(f ⊗ g) ∇M⊗N = ∇M ′⊗N ′ (f ⊗ g).
In a similar way, if (M,M ) and (N, N ) are left H-comodules the morphism
∇′M⊗N
= (εH ⊗M ⊗N) M⊗M : M ⊗N →M ⊗N
is idempotent. We denote by M ⊙N the image of ∇′M⊗N
and by p′M,N
: M ⊗N →M ⊙N ,
i′M,N
: M ⊙N →M ⊗N the morphisms such that i′M,N
p′M,N
= ∇′M⊗N
and p′M,N
i′M,N
=
idM⊙N . Using the definition of ⊙ we obtain that the object M ⊙N is a left H-comodule with
coaction M⊙N = (H ⊗ p′M,N
) M⊗N i′M,N
: M ⊙N → H ⊗ (M ⊙N). If f : M →M ′ and
g : N → N ′ are morphisms of left H-comodules then (f ⊗ g) ∇′M⊗N
= ∇′M ′⊗N ′ (f ⊗ g).
Let (M,ϕM ), (N,ϕN ), (P,ϕP ) be left H-modules. Then the following equalities hold
(Lemma 1.7 of [3]):
ϕM⊗N (H ⊗∇M⊗N ) = ϕM⊗N ,
∇M⊗N ϕM⊗N = ϕM⊗N = ϕM⊗N ∇M⊗N ,
(iM,N ⊗ P ) ∇(M×N)⊗P (pM,N ⊗ P ) = (M ⊗ iN,P ) ∇M⊗(N×P ) (M ⊗ pN,P ),
(M⊗iN,P )∇M⊗(N×P )(M⊗pN,P ) = (∇M⊗N⊗P )(M⊗∇N⊗P ) = (M⊗∇N⊗P )(∇M⊗N⊗P ).
Furthermore, by a similar calculus, if (M,M ), (N, N ), (P, P ) be left H-comodules we
have
(H ⊗∇′M⊗N
) M⊗N = M⊗N ,
M⊗N ∇′M⊗N
= M⊗N = ∇′M⊗N
M⊗N ,
158
(i′M,N
⊗ P ) ∇′(M⊙N)⊗P
(p′M,N
⊗ P ) = (M ⊗ i′N,P
) ∇′M⊗(N⊙P )
(M ⊗ p′N,P
),
(M⊗i′N,P
)∇′M⊗(N⊙P )
(M⊗p′N,P
) = (∇′M⊗N
⊗P )(M⊗∇′N⊗P
) = (M⊗∇′N⊗P
)(∇′M⊗N
⊗P ).
Yetter-Drinfeld modules over finite dimensional weak Hopf algebras over fields have been
introduced by Bohm in [9]. It is shown in [9] that the category of finite dimensional Yetter-
Drinfeld modules is monoidal and in [18] it is proved that this category is isomorphic to the
category of finite dimensional modules over the Drinfeld double. In [12], the results of [18] are
generalized, using duality results between entwining structures and smash product structures,
and more properties are given.
Definition 1.6. Let H be a weak Hopf algebra. We shall denote by H
HYD the category of
left-left Yetter-Drinfeld modules over H. That is, M = (M,ϕM , M ) is an object in H
HYD if
(M,ϕM ) is a left H-module, (M,M ) is a left H-comodule and
(b1) (µH ⊗M) (H ⊗ cM,H) ((M ϕM ) ⊗H) (H ⊗ cH,M ) (δH ⊗M)
= (µH ⊗ ϕM ) (H ⊗ cH,H ⊗M) (δH ⊗ M ).
(b2) (µH ⊗ ϕM ) (H ⊗ cH,H ⊗M) ((δH ηH) ⊗ M ) = M .
Let M , N in H
HYD. The morphism f : M → N is a morphism of left-left Yetter-Drinfeld
modules if f ϕM = ϕN (H ⊗ f) and (H ⊗ f) M = N f .
Note that if (M,ϕM , M ) is a left-left Yetter-Drinfeld module then (b2) is equivalent to
(b3) ((εH µH) ⊗ ϕM ) (H ⊗ cH,H ⊗M) (δH ⊗ M ) = ϕM .
and we have the identity ϕM (ΠL
H⊗M) M = idM .
The conditions (b1) and (b2) of the last definition can also be restated (see Proposition
2.2 of [12]) in the following way: suppose that (M,ϕM ) ∈ | HC| and (M,M ) ∈ | HC|, then
M is a left-left Yetter-Drinfeld module if and only if
M ϕM = (µH ⊗M) (H ⊗ cM,H)
(((µH ⊗ ϕM ) (H ⊗ cH,H ⊗M) (δH ⊗ M )) ⊗ λH) (H ⊗ cH,M ) (δH ⊗M).
Moreover, the following Proposition, proved in [4], guaranties the equality between the
morphisms ∇M⊗N and ∇′M⊗N
defined in 1.5 for all M,N ∈ | H
HYD|.
Proposition 1.7. Let H be a weak Hopf algebra. Let (M,ϕM , M ) and (N,ϕN , N ) be left-left
Yetter-Drinfeld modules over H. Then we have the following assertions.
(i) ∇M⊗N = ((ϕM (ΠL
H⊗M) cM,H) ⊗N) (M ⊗ N ).
(ii) ∇′M⊗N
= (M ⊗ ϕN ) (((M ⊗ ΠR
H) cH,M M ) ⊗N).
159
(iii) ∇M⊗N = ∇′M⊗N
.
(iv) ∇M⊗H = ((ϕM (ΠL
H⊗M) cM,H) ⊗H) (M ⊗ δH).
(v) ∇′M⊗H
= (M ⊗ µH) (((M ⊗ ΠR
H) cH,M M ) ⊗H).
(vi) ∇M⊗H = ∇′M⊗H
.
1.8. It is a well know fact that, if the antipode of a weak Hopf algebra H is invertible, H
HYD
is a non-strict braided monoidal category. In the following lines we give a brief resume of the
braided monoidal structure that we can construct in the category H
HYD (see Proposition 2.7
of [18] for modules over a field K or Theorem 2.6 of [12] for modules over a commutative
ring).
For two left-left Yetter-Drinfeld modules (M,ϕM , M ), (N,ϕN , N ) the tensor product is
defined as object as the image of ∇M⊗N (see 1.5). As a consequence, by (iii) of Proposition
1.7, M ×N = M ⊙N and this object is a left-left Yetter-Drinfeld module with the following
action and coaction:
ϕM×N = pM,N ϕM⊗N (H ⊗ iM,N ), M×N = (H ⊗ pM,N) M⊗N iM,N .
The base object is HL = Im(ΠL
H) or, equivalently, the equalizer of δH and ζ1
H= (H ⊗
ΠL
H) δH (see (9)) or the equalizer of δH and ζ2
H= (H ⊗ Π
R
H) δH . The structure of left-left
Yetter-Drinfeld module for HL is the one derived of the following morphisms
ϕHL= pL µH (H ⊗ iL), HL
= (H ⊗ pL) δH iL.
where pL : H → HL and iL : HL → H are the morphism such that ΠL
H= iL pL and
pL iL = idHL.
The unit constrains are:
lM = ϕM (iL ⊗M) iHL,M : HL ×M →M,
rM = ϕM cM,H (M ⊗ (ΠL
H iL)) iM,HL
: M ×HL →M.
These morphisms are isomorphisms with inverses:
l−1
M= pHL,M (pL ⊗ ϕM ) ((δH ηH) ⊗M) : M → HL ×M,
r−1
M= pM,HL
(ϕM ⊗ pL) (H ⊗ cH,M ) ((δH ηH) ⊗M) : M →M ×HL.
If M , N , P are objects in the category H
HYD, the associativity constrains are defined by
aM,N,P = p(M×N),P (pM,N ⊗ P ) (M ⊗ iN,P ) iM,(N×P ) : M × (N × P ) → (M ×N) × P
where the inverse is the morphism
a−1
M,N,P= aM,N,P = pM,(N×P )(M⊗pN,P )(iM,N⊗P )i(M×N),P : (M×N)×P →M×(N×P ).
160
If γ : M →M ′ and φ : N → N ′ are morphisms in the category, then
γ × φ = pM ′×N ′ (γ ⊗ φ) iM,N : M ×N →M ′ ×N ′
is a morphism in H
HYD and (γ′ × φ′) (γ × φ) = (γ′ γ)× (φ′ φ), where γ′ : M ′ →M ′′ and
φ′ : N ′ → N ′′ are morphisms in H
HYD.
Finally, the braiding is
τM,N = pN,M tM,N iM,N : M ×N → N ×M
where tM,N = (ϕN ⊗M) (H ⊗ cM,N ) (M ⊗N) : M ⊗N → N ⊗M . The morphism τM,N
is a natural isomorphism with inverse:
τ−1
M,N= pM,N t′
M,N iN,M : N ×M →M ×N
where t′M,N
= cN,M (ϕN ⊗M) (cN,H ⊗M) (N ⊗ λ−1
H⊗M) (N ⊗ M ).
2 Projections, quantum groupoids and crossed products
In this section we give basic properties of quantum groupoids with projection. The material
presented here can be found in [1] and [2]. For example, in Theorem 2.2 we will show that if
H, B are quantum groupoids in C and g : B → H is a quantum groupoid morphism such that
there exist a coalgebra morphism f : H → B verifiying g f = idH and f ηH = ηB then,
it is possible to find an object BH , defined by an equalizer diagram an called the algebra of
coinvariants, morphisms ϕBH: H ⊗BH → BH , σBH
: H ⊗H → BH and an isomorphism of
algebras and comodules bH : B → BH × H being BH ×H a subobject of BH ⊗ H with its
algebra structure twisted by the morphism σBH. Of course, the multiplication in BH ×H is
a generalization of the crossed product and in the Hopf algebra case the Theorem 2.2 is the
classical and well know result obtained by Blattner, Cohen and Montgomery in [6].
The following Proposition is a generalization to the quantum groupoid setting of classic
result obtained by Radford in [23].
Proposition 2.1. Let H, B be quantum groupoids in C. Let g : B → H be a morphism of
quantum groupoids and f : H → B be a morphism of coalgebras such that g f = idH . Then
the following morphism is an idempotent in C:
qB
H= µB (B ⊗ (λB f g)) δB : B → B.
Proof. See Proposition 2.1 of [2].
As a consequence of this proposition, we obtain that there exist an epimorphism pB
H, a
monomorphism iBH
and an object BH such that the diagram
-
HHHHj >
B B
BH
qB
H
pB
HiBH
161
commutes and pB
H iB
H= idBH
. Moreover, we have that
- --BH B ⊗H
iBH
(B ⊗ g) δB
(B ⊗ (ΠL
H g)) δB
B
is an equalizer diagram.
Now, let ηBHand µBH
be the factorizations, through the equalizer iBH
, of the morphisms
ηB and µB (iBH⊗ iB
H). Then (BH , ηBH
= pB
H ηB , µBH
= pB
H µB (iB
H⊗ iB
H)) is an algebra
in C.
On the other hand, by Proposition 2.4 of [2] we have that there exists an unique morphism
ϕBH: H ⊗ BH → BH such that iB
H ϕBH
= yB where yB : H ⊗ BH → B is the morphism
defined by yB = µB (B ⊗ (µB cB,B)) (f ⊗ (λB f)⊗B) (δH ⊗ iBH
). The morphism ϕBH
satisfies:
ϕBH= pB
H µB (f ⊗ iB
H),
ϕBH (ηH ⊗BH) = idBH
,
ϕBH (H ⊗ ηBH
) = ϕBH (ΠL
H⊗ ηBH
),
µBH (ϕBH
⊗BH) (H ⊗ ηBH⊗BH) = ϕBH
(ΠL
H⊗BH),
ϕBH (H ⊗ µBH
) = µBH (ϕBH
⊗ ϕBH) (H ⊗ cH,BH
⊗BH) (δH ⊗BH ⊗BH),
µBH cBH ,BH
((ϕBH (H ⊗ ηBH
)) ⊗BH) = ϕBH (Π
L
H⊗BH).
and, if f is an algebra morphism, (BH , ϕBH) is a left H-module (Proposition 2.5 of [1]).
Moreover, in this setting, there exists an unique morphism σBH: H ⊗H → BH such that
iBH σBH
= σB where σB : H ⊗H → B is the morphism defined by:
σB = µB ((µB (f ⊗ f)) ⊗ (λB f µH)) δH⊗H .
Then, as a consequence, we have the equality σBH= pB
H σB (Proposition 2.6, [2]).
Now let ωB : BH ⊗H → B be the morphism defined by ωB = µB (iBH⊗ f). If we define
ω′B
: B → BH ⊗ H by ω′B
= (pB
H⊗ g) δB we have ωB ω′
B= idB . Then, the morphism
ΩB = ω′B ωB : BH ⊗H → BH ⊗H is idempotent and there exists a diagram
-Z
ZZ
ZZ~
3 Z
ZZZ~
?
BH ⊗H BH ⊗H
B
BH ×H
ωBω′
B
rB sB
ΩB
bB
where sB rB = ΩB, rB sB = idBH×H , bB = rB ω′B.
162
It is easy to prove that the morphism bB is an isomorphism with inverse b−1
B= ωB sB .
Therefore, the object BH×H is an algebra with unit and product defined by ηBH×H = bBηB ,
µBH×H = bB µB (b−1
B⊗ b−1
B) respectively. Also, BH × H is a right H-comodule where
ρBH×H = (bB ⊗H) (B⊗ g) δB b−1
B. Of course, with these structures bB is an isomorphism
of algebras and right H-comodules being ρB = (B ⊗ g) δB .
On the other hand, we can define the following morphisms:
ηBH♯σBHH : K → BH×H, µBH♯σBH
H : BH×H⊗BH×H → BH×H, ρBH♯σBHH : BH → BH×H⊗H
where
ηBH♯σBHH = rB (ηBH
⊗ ηH),
µBH♯σBHH = rB (µBH
⊗H) (µBH⊗ σBH
⊗ µH) (BH ⊗ ϕBH⊗ δH⊗H)
(BH ⊗H ⊗ cH,BH⊗H) (BH ⊗ δH ⊗BH ⊗H) (sB ⊗ sB),
ρBH♯σBHH = (rB ⊗H) (BH ⊗ δH) sB.
Finally, if we denote by BH♯σBHH (the crossed product of BH and H) the triple
(BH ×H, ηBH ♯σBHH , µBH♯σBH
H)
we have the following theorem.
Theorem 2.2. Let H, B be quantum groupoids in C. Let g : B → H be a morphism of
quantum groupoids and f : H → B be a morphism of coalgebras such that g f = idH and
f ηH = ηB. Then, BH♯σBHH is an algebra, (BH ×H, ρBH♯σBH
H) is a right H-comodule and
bB : B → BH♯σBHH is an isomorphism of algebras and right H-comodules.
Proof: The proof of this Theorem is a consequence of the following identities (see Theorem
2.8 of [2] for the complete details)
ηBH♯σBHH = ηBH×H , µBH♯σBH
H = µBH×H , ρBH♯σBHH = ρBH×H .
Remark 2.3. We point out that if H and B are Hopf algebras, Theorem 2.2 is the result
obtained by Blattner, Cohen and Montgomery in [6]. Moreover, if f is an algebra morphism,
we have σBH= εH ⊗ εH ⊗ ηBH
and then BH♯σBHH is the smash product of BH and H,
denoted by BH♯H. Observe that the product of BH♯H is
µBH♯H = (µBH⊗ µH) (BH ⊗ ((ϕBH
⊗H) (H ⊗ cH,BH) (δH ⊗BH)) ⊗H)
Let H, B be quantum groupoids in C. Let g : B → H, f : H → B be morphisms of
quantum groupoids such that g f = idH . In this case σB = ΠB
L f µH and then, using
µB (ΠL
B⊗B) δB = idB , we obtain
µBH♯σBHH = rB (µBH
⊗µH) (BH ⊗ ((ϕBH⊗H) (H ⊗ cH,BH
) (δH ⊗BH))⊗H) (sB ⊗sB)
163
As a consequence, for analogy with the Hopf algebra case, when σB = ΠB
L f µH , we
will denote the triple BH♯σBHH by BH♯H (the smash product of BH and H).
Therefore, if f and g are morphisms of quantum groupoids, we have the following partic-
ular case of 2.2.
Corollary 2.4. Let H, B be quantum groupoids in C. Let g : B → H, f : H → B be
morphisms of quantum groupoids such that g f = idH . Then BH♯H is an algebra, (BH ×
H, ρBH♯H) is a right H-comodule and bB : B → BH♯H is an isomorphism of algebras and
right H-comodules.
In a similar way we can obtain a dual theory. The arguments are similar to the ones
used previously in this section, but passing to the opposite category. Let H, B be quantum
groupoids in C. Let h : H → B be a morphism of quantum groupoids and t : B → H be a
morphism of algebras such that t h = idH and εH t = εB . The morphism kB
H: B → B
defined by
kB
H= µB (B ⊗ (h t λB)) δB
is idempotent in C and, as a consequence, we obtain that there exist an epimorphism lBH
, a
monomorphism nB
Hand an object BH such that the diagram
-
HHHHj >
B B
BH
kB
H
lBH
nB
H
commutes and lBH nB
H= idBH . Moreover, using the next coequalizer diagram in C
-- -
µB (B ⊗ h)
µB (B ⊗ (h ΠL
H))
lBH
B ⊗H B BH
it is possible to obtain a coalgebra structure for BH . This structure is given by
(BH , εBH = εB nB
H, δ
BH = (lBH⊗ lB
H) δB nB
H)).
Let yB : B → H ⊗BH be the morphism defined by:
yB = (µH ⊗ lBH
) (t⊗ (t λB) ⊗B) (B ⊗ (cB,B δB)) δB .
The morphism yB verifies that yB µB (B ⊗ h) = yB µB (B ⊗ (ΠL
B h)) and then,
there exists an unique morphism rBH : BH → H ⊗BH such that r
BH lBH
= yB.
Moreover the morphism BH satisfies:
BH = (t⊗ lB
H) δB nB
H,
164
(εH ⊗BH) BH = idBH ,
(H ⊗ εBH ) BH = (ΠL
H⊗ εBH ) BH ,
(H ⊗ εBH ⊗BH) (
BH ⊗BH) δBH= (ΠL
H⊗BH)
BH
(H ⊗ δBH )
BH = (µH ⊗BH ⊗BH) (H ⊗ cBH ,H
⊗BH) (BH ⊗
BH ) δBH ,
(((H ⊗ εBH ) BH ) ⊗BH) cBH ,BH δBH = (ΠL
H⊗BH) BH ,
and, if t is a morphism of quantum groupoids, (BH , BH ) is a left H-comodule. Let γB : B →
H ⊗H be the morphism defined by
γB = µH⊗H (((t⊗ t) δB) ⊗ (δH t λB)) δB .
The morphism γB verifies that γB µB (B ⊗ h) = γB µB (B ⊗ (ΠL
B h)) and then,
there exists an unique morphism γBH : BH → H ⊗H such that γBH lBH
= γB .
It is not difficult to see that the morphism ΥB : BH ⊗H → BH ⊗H defined by
ΥB = ′BB ,
being B = µB (nB
H⊗ h) and ′
B= (lB
H⊗ t) δB , is idempotent and there exists a diagram
-Z
ZZ
ZZ~
3 Z
ZZZ~
?
BH ⊗H BH ⊗H
B
BH ⊡H
B′
B
uB vB
ΥB
dB
where vB uB = ΥB, uB vB = idBH⊡H , dB = uB ′B. Moreover, dB is an isomorphism
with inverse d−1
B= B vB and the object BH ⊡H is a coalgebra with counit and coproduct
defined by
εBH⊡H
= εB d−1
B, δ
BH⊡H= (dB ⊗ dB) δB d−1
B
respectively.
Also, BH ⊡H is a right H-module where
ψBH⊡H
= dB µB (d−1
B⊗ h).
With these structures dB is an isomorphism of coalgebras and right H-modules being
ψB = µB (B ⊗ h). Finally, we define the morphisms:
εBH⊖γBH
H : BH ⊡H → K, δBH⊖γBH
H : BH ⊡H → BH ⊡H ⊗BH ⊡H,
ψBH⊖γ
BHH
: BH ⊡H ⊗H → BH ⊡H
where
165
εBH⊖γBH
H = (εBH ⊗ εH) vB ,
δBH⊖γ
BHH
= (uB ⊗ uB) (BH ⊗ µH ⊗BH ⊗H) (BH ⊗H ⊗ cBH ,H
⊗H)
(BH ⊗ BH ⊗ µH⊗H) (δBH ⊗ γBH ⊗ δH) (δBH ⊗H) vB ,
ψBH⊖γBH
H = uB (BH ⊗ µH) (vB ⊗H).
If we denote by BH ⊖γBH
H (the crossed coproduct of BH and H) the triple
(BH ⊡H, εBH⊖γ
BHH, δ
BH⊖γBH
H),
we have the following theorem:
Theorem 2.5. Let H, B be quantum groupoids in C. Let h : H → B be a morphism of
quantum groupoids and t : B → H be a morphism of algebras such that t h = idH and
εH t = εB. Then, BH ⊖γBH
H is a coalgebra, (BH ⊡ H,ψBH⊖γ
BHH
) is a right H-module
and dB : B → BH ⊖γBH
H is an isomorphism of coalgebras and right H-modules.
Remark 2.6. In the Hopf algebra case (H and B Hopf algebras) Theorem 2.5 is the dual
of the result obtained by Blattner, Cohen and Montgomery. In this case, if t is an algebra-
coalgebra morphism, we have γBH = ε
BH ⊗ ηH ⊗ ηH and then BH ⊖γBH
H is the smash
coproduct of BH and H, denoted by BH ⊖H. In BH ⊖H the coproduct is
δBH⊖H
= (BH ⊗ ((µH ⊗BH) (H ⊗ cBH ,H
) (BH ⊗H)) ⊗H) (δ
BH ⊗ δH).
If t is a morphism of quantum groupoids we have γB = δH ΠL
Ht and then the expression
of δBH⊖γBH
H is:
δBH⊖γ
BHH
= (uB ⊗uB)(BH ⊗((µH ⊗BH)(H⊗cBH ,H
)(BH ⊗H))⊗H)(δ
BH ⊗δH)vB .
As a consequence, for analogy with the Hopf algebra case, when γB = δH ΠH
L t, we will
denote the triple BH ⊖γBH
H by BH ⊖H (the smash coproduct of BH and H).
Therefore, if h and t are morphisms of quantum groupoids, we have:
Corollary 2.7. Let H, B be quantum groupoids in C. Let t : B → H, h : H → B be
morphisms of quantum groupoids such that t h = idH . Then, BH ⊖ H is a coalgebra,
(BH ⊡H,ψBH⊖H
) is a right H-module and dB : B → BH ⊖H is an isomorphism of coalgebras
and right H-modules.
166
3 Quantum groupoids, projections and Hopf algebras inHHYD
In this section we give the connection between projection of quantum groupoids an Hopf
algebras in the category H
HYD. The results presented here can be found in [3].
Suppose that g : B → H and f : H → B are morphisms of weak Hopf algebras such that
g f = idH . Then qB
H= kB
Hand therefore BH = BH , pB
H= lB
Hand iB
H= nB
H. Thus
- --BH B B ⊗H
iBH
(B ⊗ g) δB
(B ⊗ (ΠL
H g)) δB
is an equalizer diagram and
-- -
µB (B ⊗ f)
µB (B ⊗ (f ΠL
H))
pH
BB ⊗H B BH
is a coequalizer diagram.
Then (BH , ηBH= pB
H ηB , µBH
= pB
H µB (iB
H⊗ iB
H)) is an algebra in C, (BH , εBH
=
εB iBH, δBH
= (pB
H⊗ pB
H) δB iB
H)) is a coalgebra in C, (BH , ϕBH
) is a left H-module and
(BH , BH) is a left H-comodule.
Also, ωB = B , ω′B
= ′B
and then BH × H = BH ⊡ H. Moreover, the morphism
ΩB = ω′B ωB admits a new formulation. Note that by the usual arguments in the quantum
groupoid calculus, we have
ΩB = (pB
H⊗ µH) (µB ⊗H ⊗ g) (B ⊗ cH,B ⊗B) (((B ⊗ g) δB iB
H) ⊗ (δB f))
= (pB
H⊗µH)(µB⊗H⊗H)(B⊗cH,B⊗H)(((B⊗(Π
R
Hg))δBiB
H)⊗((f⊗H)δH )))
= (pB
H⊗ εH ⊗H) (µB⊗H ⊗H) (((B ⊗ g) δB iB
H) ⊗ ((f ⊗ δH) δH)))
= (pB
H⊗ (εH g) ⊗H) (µB⊗B ⊗H) (δB ⊗ δB ⊗H) (iB
H⊗ ((f ⊗H) δH))
= ((pB
H µB) ⊗H) (iB
H⊗ ((f ⊗H) δH))
= ((pB
H µB (B ⊗ qB
H)) ⊗H) (iB
H⊗ ((f ⊗H) δH))
= (pB
H⊗H) ((µB (B ⊗ (ΠL
B f)) ⊗H) (iB
H⊗ δH)
= (pB
H⊗H) ((µB cB,B ((ΠL
B f) ⊗ iB
H)) ⊗H) (cBH ,H ⊗H) (BH ⊗ δH)
= ((pB
H iB
H ϕBH
(ΠL
H⊗BH)) ⊗H) (cBH ,H ⊗H) (BH ⊗ δH)
= (ϕBH⊗H) (cBH ,H ⊗H) (BH ⊗ Π
L
H⊗H) (BH ⊗ δH)
= (ϕBH⊗ µH) (H ⊗ cH,BH
⊗H) ((δH ηH) ⊗BH ⊗H).
= ∇BH⊗H .
167
Therefore, the object BH × H is the tensor product of BH and H in the representation
category of H, i.e. the category of left H-modules, studied in [8] and [21].
Proposition 3.1. Let g : B → H and f : H → B be morphisms of quantum groupoids such
that g f = idH . Then, if the antipode of H is an isomorphism, (BH , ϕBH, BH
) belongs toH
HYD.
Proof: In Proposition 2.8 of [1] we prove that (BH , ϕBH, BH
) satisfy
(µH ⊗BH) (H ⊗ cBH ,H) ((BH ϕBH
) ⊗H) (H ⊗ cH,BH) (δH ⊗BH)
= (µH ⊗BH)(H⊗cBH ,H)(µH ⊗ϕBH⊗H)(H⊗cH,H ⊗BH ⊗H)(δH ⊗BH
⊗ΠR
H)
(H ⊗ cH,BH) (δH ⊗BH).
Moreover, the following identity
(µH ⊗BH) (H ⊗ cBH ,H) (µH ⊗ ϕBH⊗H) (H ⊗ cH,H ⊗BH ⊗H) (δH ⊗ BH
⊗ ΠR
H)
(H ⊗ cH,BH) (δH ⊗BH)
= (µH ⊗ ϕBH) (H ⊗ cH,H ⊗ (ϕBH
((ΠL
H Π
R
H) ⊗BH) BH
)) (δH ⊗ BH).
is true because BH is a left H-module and a left H-comodule. Then, using the identity
ϕBH ((Π
L
H Π
R
H) ⊗BH) BH
= idBH
we prove (b1). The prove for (b2) is easy and we leave the details to the reader.
3.2. As a consequence of the previous proposition we obtain ∇BH⊗BH= ∇′
BH⊗BHand
∇BH⊗H = ∇′BH⊗H
= ΩB.
3.3. Let g : B → H and f : H → B be morphisms of quantum groupoids such that gf = idH .
Put uBH= pB
H f iL : HL → BH and eBH
= pL g iBH
: BH → HL. This morphisms belong
to H
HYD and we have the same for mBH×BH
: BH ×BH → BH defined by
mBH×BH= µBH
iBH ,BH
and ∆BH: BH → BH ×BH defined by ∆BH
= pBH ,BH δBH
.
Then, we have the following result.
Proposition 3.4. Let g : B → H and f : H → B be morphisms of quantum groupoids such
that g f = idH . Then, if the antipode of H is an isomorphism, we have the following:
(i) (BH , uBH,mBH
) is an algebra in H
HYD.
(ii) (BH , eBH,∆BH
) is a coalgebra in H
HYD.
168
Proof: See Proposition 2.6 in [3].
3.5. Let g : B → H and f : H → B be morphisms of weak Hopf algebras such that gf = idH .
Let ΘB
Hbe the morphism ΘB
H= ((f g)∧λB) iB
H: BH → B. Following Proposition 2.9 of [1]
we have that (B⊗ g) δB ΘB
H= (B⊗ (ΠL
H g)) δB ΘB
Hand, as a consequence, there exists
an unique morphism λBH: BH → BH such that iB
H λBH
= ΘB
H. Therefore, λBH
= pB
H ΘB
H
and λBHbelongs to the category of left-left Yetter-Drinfeld modules.
The remainder of this section will be devoted to the proof of the main Theorem of this
paper.
Theorem 3.6. Let g : B → H and f : H → B be morphisms of weak Hopf algebras
satisfying the equality g f = idH and suppose that the antipode of H is an isomorphism.
Let uBH, mBH
, eBH, ∆BH
, λBHbe the morphisms defined in 3.3 and 3.5 respectively. Then
(BH , uBH,mBH
, eBH,∆BH
, λBH) is a Hopf algebra in the category of left-left Yetter-Drinfeld
modules.
Proof: By Proposition 3.4 we know that (BH , uBH,mBH
) is an algebra and (BH , eBH,∆BH
)
is a coalgebra in H
HYD.
First we prove that mBHis a coalgebra morphism. That is:
(c1) ∆BHmBH
= (mBH×mBH
) aBH ,BH ,BH×BH (BH × a−1
BH ,BH ,BH)
(BH×(τBH ,BH×BH))(BH×aBH ,BH ,BH
)a−1
BH ,BH ,BH×BH(∆BH
×∆BH),
(c2) eBHmBH
= lHL (eBH
× eBH).
Indeed:
(mBH×mBH
) aBH ,BH ,BH×BH (BH × a−1
BH ,BH ,BH) (BH × (τBH ,BH
×BH))
(BH × aBH ,BH ,BH) a−1
BH ,BH ,BH×BH (∆BH
× ∆BH)
= pBH ,BH (µBH
⊗ µBH) (BH ⊗ iBH ,BH
⊗BH) (∇BH⊗(BH×BH) ⊗BH)
(BH⊗∇(BH×BH)⊗BH)(BH⊗(pBH ,BH
tBH ,BHiBH ,BH
)⊗BH)(BH⊗∇(BH×BH)⊗BH)
(∇BH⊗(BH×BH) ⊗BH) (BH ⊗ pBH ,BH⊗BH) (δBH
⊗ δBH) iBH ,BH
= pBH ,BH (µBH
⊗ µBH) (BH ⊗ (∇BH⊗BH
tBH ,BH∇BH⊗BH
)⊗BH) (δBH⊗ δBH
)
iBH ,BH
= pBH ,BH (µBH
⊗ µBH) (BH ⊗ tBH ,BH
⊗BH) (δBH⊗ δBH
) iBH ,BH
= pBH ,BH δBH
µBH iBH ,BH
169
= ∆BHmBH
.
In the last computations, the first and the second equalities follow from Lemma 1.7 of
[3] and by µBH ∇BH⊗BH
= µBH, ∇BH⊗BH
δBH= δBH
. In the third one we use the
following result: if M is a left-left Yetter-Drinfeld module then tM,M ∇M⊗M = tM,M ,
∇M⊗M tM,M = tM,M . The fourth equality follows from Proposition 2.9 of [1] and, finally,
the fifth one follows by definition.
On the other hand,
lHL (eBH
× eBH)
= pL µH (iL ⊗ iL) ∇HL⊗HL (pL ⊗ pL) ((g iB
H) ⊗ (g iB
H)) iBH ,BH
= pL µH ((ΠL
H g iB
H) ⊗ (ΠL
H g iB
H)) iBH ,BH
= pL µH ((g qB
H iB
H) ⊗ (g qB
H iB
H)) iBH ,BH
= pL µH ((g iBH
) ⊗ (g iBH
)) iBH ,BH
= pL g iBH µBH
iBH ,BH
= eBHmBH
.
The first equality follows from definition, the second one from
pL µH (iL ⊗ iL) ∇HL⊗HL (pL ⊗ pL) = pL µH (ΠL
H⊗ ΠL
H)
and the third one from ΠL
H g = g qB
H. Finally, the fourth one follows from the idempotent
character of qB
H, the fifth one from the properties of g and the definition of µBH
and the sixth
one from definition.
To finish the proof we only need to show
mBH (λBH
×BH) ∆BH= lBH
(eBH× uBH
) r−1
BH= mBH
(BH × λBH) ∆BH
.
We begin by proving lBH (eBH
× uBH) r−1
BH= uBH
eBH. Indeed:
lBH (eBH
× uBH) r−1
BH
= pB
HµB(f⊗B)(iL⊗i
B
H)∇HL⊗BH
(pL⊗pB
H)(g⊗f)(iB
H⊗iL)∇BH⊗HL
(pB
H⊗pL)
((µB (f ⊗ iBH
)) ⊗H) (H ⊗ cH,BH) ((δH ηH) ⊗BH)
= pB
H µB ((ΠL
B∧ΠL
B)⊗ΠL
B) ((f g qB
H)⊗ (µB (ΠL
B⊗ (f g ΠL
B)))) (δB ⊗B)
δB iBH
= pB
H µB ((ΠL
B f g qB
H) ⊗ (f ΠL
H g ΠL
B)) δB iB
H
170
= pB
H f µH (ΠL
H⊗ ΠL
H) δH g iB
H
= pB
H f ΠL
H g iB
H
= uBH eBH
.
The first equality follows from definition, the second one from
((µB (f ⊗ iBH
)) ⊗H) (H ⊗ cH,BH) ((δH ηH) ⊗BH) = (B ⊗ (g ΠL
B)) δB iB
H,
(iBH⊗ iL) ∇BH⊗HL
(pB
H⊗ pL) = (qB
H⊗ (ΠL
H g µB)) (B ⊗ ΠL
B⊗ f) (δB ⊗H)
and
(iL ⊗ iBH
) ∇HL⊗BH (pL ⊗ pB
H) = (ΠL
H g) ⊗ (qB
H µB)) (B ⊗ ΠL
B⊗B) ((δB f) ⊗B).
In the third one we use ΠL
B∧ΠL
B= ΠL
B. The fourth one follows from ΠL
Hg = g qB
Hand from
the idempotent character of ΠL
H. Finally, in the fifth one we apply (75) for ΠL
H∧ ΠL
H= ΠL
H.
On the other hand,
mBH (λBH
×BH) ∆BH
= µBH ∇BH⊗BH
(λBH⊗BH) ∇BH⊗BH
δBH
= µBH (λBH
⊗BH) δBH
= ((εBH µBH
) ⊗BH) (BH ⊗ tBH ,BH) ((δBH
ηBH) ⊗BH)
= ((εB qB
H µB) ⊗ pB
H) ((µB (qB
H⊗ (f g)) δB) ⊗ cB,B) ((δB qB
H ηB) ⊗ iB
H)
= pB
H ΠL
B iB
H
= pB
H f ΠL
H g iB
H
= uBH eBH
.
In these computations, the first equality follows from definition, the second one from
µBH ∇BH⊗BH
= µBHand ∇BH⊗BH
δBH= δBH
, the third one from (4-1) of Proposition
2.9 of [1] and the fourth one is a consequence of the coassociativity of δB . The fifth equality
follows from µB (qB
H⊗ (f g)) δB = idB and qB
H ηB = ηB , εB qB
H= εB . In the sixth one
we use f ΠL
H g = ΠL
Band the last one follows from definition.
Finally, using similar arguments and (4-2) of Proposition 2.9 of [1] we obtain:
mBH (BH × λBH
) ∆BH
= µBH ∇BH⊗BH
(BH ⊗ λBH) ∇BH⊗BH
δBH
= µBH (λBH
⊗BH) δBH
171
= (BH ⊗ (εBH µBH
)) (tBH ,BH⊗BH) (BH ⊗ (δBH
ηBH))
= pB
H µB ((f g) ⊗ ΠR
B) δB iB
H
= pB
H µB ((f g) ⊗ (f ΠR
H g) δB iB
H
= pB
H f (idH ∧ ΠR
H) g iB
H
= pB
H f g iB
H
= pB
H f ΠL
H g iB
H
= uBH eBH
.
Finally, using the last theorem and Theorem 4.1 of [2] we obtain the complete version
of Radford’s Theorem linking weak Hopf algebras with projection and Hopf algebras in the
category of Yetter-Drinfeld modules over H.
Theorem 3.7. Let H, B be weak Hopf algebras in C. Let g : B → H and f : H → B be
morphisms of weak Hopf algebras such that g f = idH and suppose that the antipode of H is
an isomorphism. Then there exists a Hopf algebra BH living in the braided monoidal categoryH
HYD such that B is isomorphic to BH × H as weak Hopf algebras, being the (co)algebra
structure in BH ×H the smash (co)product, that is the (co)product defined in 2.3, 2.6. The
expression for the antipode of BH ×H is
λBH×H := pBH ,H (ϕBH⊗H)
(H ⊗ cH,BH) ((δH λH µH) ⊗ λBH
) (H ⊗ cBH ,H)
(BH⊗H) iBH ,H .
Acknowledgements
The authors have been supported by Ministerio de Educacion y Ciencia and by FEDER.
References
[1] J.N. Alonso Alvarez, R. Gonzalez Rodrıguez, J.M. Fernandez Vilaboa, M.P. Lopez Lopez,
E. Villanueva Novoa, Weak Hopf algebras with projection and weak smash bialgebra
structures, J. of Algebra, 269 (2003), 701-725.
[2] J.N. Alonso Alvarez, R. Gonzalez Rodrıguez, Crossed products for weak Hopf algebras
with coalgebra splitting, J. of Algebra, 281 (2004), 731-752.
172
[3] Alonso Alvarez, J.N., Fernandez Vilaboa, J.M., Gonzalez Rodrıguez, R., Yetter-Drinfeld
modules and projections of weak Hopf algebras, J. of Algebra (to appear), (2006).
[4] Alonso Alvarez, J.N., Fernandez Vilaboa, J.M., Gonzalez Rodrıguez, R., Weak Hopf al-
gebras and weak Yang-Baxter operators, preprint (2006).
[5] Y. Bespalov, Crossed modules and quantum groups in braided categories, Appl. Categ.
Struct. (2)5 (1997), 155-204.
[6] R. Blattner, M. Cohen and S. Montgomery, Crossed products and inner actions of Hopf
algebras, Transactions of the American Mathematical Society, 298(2) (1986), 671-711.
[7] G. Bohm, F. Nill, K. Szlachanyi, Weak Hopf algebras, I. Integral theory and C∗-structure,
J. of Algebra, 221 (1999), 385-438.
[8] G. Bohm, K. Szlachanyi, Weak Hopf algebras, II. Representation theory, dimensions and
the Markov trace, J. of Algebra, 233 (2000), 156-212.
[9] G. Bohm, Doi-Hopf modules over weak Hopf algebras, Comm. in Algebra, 28 (2000),
4687-4698.
[10] D. Bulacu, F. Panaite, F.V. Oystaeyen, Quasi-Hopf algebra actions and smash product,
Comm. in Algebra, 28 (2000), 631-651.
[11] D. Bulacu, E. Nauwelaerts, Radford’s biproduct for quasi-Hopf algebras and bosoniza-
tion, J. Pure Appl. Algebra, 174 (2002), 1-42.
[12] S. Caenepeel, D. Wang, Y. Yin, Yetter-Drinfeld modules over weak Hopf algebras and
the center construction, Ann. Univ. Ferrara - Sez. VII - Sc. Mat. 51 (2005), 69-98.
[13] F. Goodman, P. de la Harpe, V.F.R. Jones, Coxeter graphs and towers of algebras,
M.S.R.I. Publ. 14, Springer, Heilderberg, (1989).
[14] T. Hayashi, Face algebras I. A generalization of quantum group theory. J. Math. Soc.
Japan, 50, (1998), 293-315.
[15] K. Kassel, Quantum groups, G.T.M. 155, Spinger Verlag, 1995.
[16] S. Majid, Cross products by braided groups and bosonization, J. of Algebra, 163 (1994),
165-190.
[17] S. Majid, Quantum double for quasi-Hopf algebras, Lett. Math. Phys., 45 (19948), 1-9.
[18] A. Nenciu, The center construction for weak Hopf algebras, Tsukuba J. of Math. 26
(2002), 189-204.
173
[19] D. Nikshych, L. Vainerman, A Galois Correspondence for Π1 factors, math.QA/0001020
(2000).
[20] D. Nikshych, L. Vainerman, Finite Quantum Groupoids and their applications, New
Directions in Hopf Algebras, MSRI Publications, 43 (2002), 211-262.
[21] D. Nikshych, V. Turaev and L. de Vainerman, Invariants of knots and 3-manifolds from
quantum groupoids, Topology Appl., 127 (2003), 91-123.
[22] F. Nill, Axioms of weak bialgebras, math.QA/9805104 (1998).
[23] D.E. Radford, The structure of Hopf algebras with a projection, J. of Algebra, 92 (1985),
322-347.
[24] J. Renault, A groupoid approach to C∗-algebras, Lecture Notes in Math. 793, Springer-
Verlag, 1980.
[25] T. Yamanouchi, Duality for generalized Kac algebras and characterization of finite
groupoid algebras, J. of Algebra, 163 (1994), 9-50.
174
Shifted determinants over universal enveloping
algebra
Natasha Rozhkovskaya∗
Abstract
We present a family of polynomials with coefficients in the universal enveloping algebra.
These polynomials are shifted analogues of a determinant of a certain non-commutative
matrix, labeled by irreducible representations of gln(C). We show plethysm relation with
Capelli polynomials and compute the polynomials explicitly for gl2(C).
Keywords: Casimir element, universal enveloping algebra, irreducible representation, deter-
minant, characteristic polynomial, Capelli polynomial, shifted symmetric functions.
1 Introduction
Let g = gln(C) with the standard basis Eij, let Vλ be an irreducible representation of g.
The matrix Ωλ, defined by
Ωλ =∑
i,j=1, ..., n
Eij ⊗ πλ(Eji),
naturally appears in many problems of representation theory. We will call it braided Casimir
element.
Consider the symmetric algebra S(g) of g. The algebra S(g) is commutative. So if we
think of Ωλ as an element of S(g) ⊗ EndVλ, the determinant pλ(u) = det(Ωλ − u) is a well-
defined polynomial. The coefficients of Dλ(u) are invariant under the ajoint action of g on
S(g). The determinant also serves as a characteristic polynomial for Ωλ (namely, pλ(Ωλ) = 0).
Now let us consider Ωλ as a matrix with coefficients in the universal enveloping algebra
U(g). This is non-commutative algebra, and it can be viewed as a deformation of S(g).
Due to B. Kostant’s theorem [6], the matrix Ωλ (now as a matrix with coefficients in U(g)),
satisfies a characteristics equation with coefficients in the center of the universal enveloping
algebra. We also define in Section 2 a (shifted) analogue of determinant Ω(λ). Thus, we
have two deformations of the polynomial Dλ(u): a shifted determinant and a characteristic
polynomial. It is well-known, that in case of Vλ - vector representation of gl(n, C), both
deformations coincide: the shifted determinant is a characteristic polynomial of Ωλ. But
∗Department of Mathematics, Harvard Univeristy. E-mail: [email protected]
175
it turns out, that in general, these two deformations are different, which follows from the
results of Section 3. There we prove the centrality of shifted determinant and find the explicit
formula in the case of gl2(C).
With every new nice example of central elements it is natural to ask, how it is related
to the known ones. One of the most well-studied families of central polynomials is the set
of Capelli polynomials. These polynomials are parametrized by dominant weights of gln(C).
Their theory is developed in the series of works [5], [8],[11],[12],[13], etc. Again, the shifted
determinant coincides with a Capelli polynomial only in the case of vector representations,
when λ = (1). In Section 5 we prove that there is a plethysm-like relation between Capelli
polynomials and shifted determinants.
In Section 2 we define shifted determinant for a non-commutative matrix. In Section 3
we study the case of gl2(C). In Section 4 we discuss relations with characteristic polynomials
and state two conjectures about centrality of shifted determinants in general. In Section 5 we
give a formula for relation with Capelli polynomials.
Acknowledgements. It is my pleasure to express my gratitude to A. Molev and A. Ram for
interesting and valuable discussions. I would like to thank also IHES and IHP for their
hospitality, and L. V. Zharkova for special support.
2 Definitions
Let g = gln(C) with the universal enveloping algebra U(g). Denote by Z(g) the center of
U(g). Fix the basis Eij of gln(C), which consists of standard unit matrices.
Let λ = (λ1, . . . , λn) with λi − λi+1 ∈ Z+, for i = 1, . . . , n − 1, be a dominant weight
of gln(C). Denote by πλ be the corresponding irreducible rational gln(C)-representation, and
by Vλ the space of this representation. We assume that dim Vλ = m + 1. Then we define an
element Ωλ of U(g) ⊗ End(Vλ), which we call braided Casimir element.
Definition.
Ωλ = Eij ⊗∑
i,j=1, ..., n
πλ(Eji).
We will think of Ωλ as a matrix of size (m + 1) × (m + 1) with coefficients in U(gln(C)).
Next we define a shifted determinant of Ωλ.
Let A be an element of A ⊗ End (Cm+1), where A is a non-commutative algebra. We
again think of A as a non-commutative matrix of size (m + 1) × (m + 1) with coefficients
Aij ∈ A.
Definition. The (column)-determinant of A is the following element of A:
det(A) =∑
σ∈Sm+1
(−1)σAσ(1)1Aσ(2)2...Aσ(m+1)(m+1) . (1)
176
Here the sum is taken over all elements σ of the symmetric group Sm+1, and (−1)σ is the
sign of the permutation σ.
Put Ωλ(u) = Ωλ + u ⊗ id. Define L as a diagonal matrix of the size (m + 1) × (m + 1) of
the form:
L = diag(m,m − 1, . . . , 0).
Definition. The shifted determinant of Ωλ(u) is the column-determinant det(Ωλ(u) − L).
We will use notation Dλ(u) for this polynomial with coefficients in U(gln(C)):
Dλ(u) = det(Ωλ(u) − L).
There is another way to define the same determinant. Let A1, . . . , As be a set of matrices
of size (m + 1) × (m + 1) with coefficients in some associative (non-commutative) algebra A.
Let µ be the multiplication in A. Consider an element of A⊗ End(Cm+1)
Λs(A1 ⊗ · · · ⊗ As) = (µ(s) ⊗ Asyms)(A1 ⊗ · · · ⊗ As),
where Asyms = 1
s!
∑
σ∈Ss(−1)σσ. By Young’s construction, the antisymmetrizer can be
realized as an element of End ((Cm+1)⊗s).
Lemma 2.1. For s = m + 1
Λm+1(A1 ⊗ · · · ⊗ Am+1) = α(A1, . . . , Am+1) ⊗ Asymm+1, (2)
where α(A1, . . . , Am+1) ∈ A,
α(A1, . . . , Am+1) =∑
σ∈Sm+1
(−1)σ [A1]σ(1),1 . . . [Am+1]σ(m+1),m+1,
and [Ak]i,j are matrix elements of Ak.
Proof. (cf [9].) Let ei, (i = 1, . . . , m+1), be a basis of V = Cm+1. Observe that Asym (m+1)
is a one-dimensional projector to
v =1
(m + 1)!
∑
σ∈Sm+1
(−1)σeσ(1) ⊗ · · · ⊗ eσ(m+1).
We apply Λm+1(A1 ⊗ · · · ⊗ Am+1) to e1 ⊗ · · · ⊗ em+1 ∈ V ⊗m+1:
Λm+1(A1 ⊗ · · · ⊗ Am+1)(e1 ⊗ · · · ⊗ em)
=∑
i1,...ik
(A1) i1, 1 . . . (Am+1) im+1, m+1 Asym (m+1) (ei1⊗ · · · ⊗ eim+1
). (3)
The vector Asym (m+1) (ei1⊗ · · · ⊗ eim+1
) 6= 0 only if all indices i1, . . . , im+1 are pairwise
distinct. In this case denote by σ be a permutation defined by σ(k) = ik. Then
Asym (m+1) (ei1⊗ · · · ⊗ eim+1
) = (−1)σ v,
177
and (3) gives
Λm+1(A1 ⊗ · · · ⊗ Am+1) (e1 ⊗ · · · ⊗ em+1) = α(A1, . . . , Am+1) v
= α(A1, . . . , Am+1)Asym (m+1) (e1 ⊗ · · · ⊗ em+1).
Corollary 2.2.
α (A, . . . , A) = det(A),
α (Ωλ(u − m), . . . , Ωλ(u)) = Dλ(u).
3 Case of gl2(C)
In this section we prove the centrality of polynomials Dλ(u) for g = gl2(C) an write them
explicitly.
The center Z(gl2(C)) of the universal enveloping algebra U(gl2(C)) is generated by two
elements:
∆1 = E11 + E22, ∆2 = (E11 − 1)E22 − E12E21.
Let λ = (λ1 ≥ λ2) be a dominant weight. Put m = λ1−λ2, d = λ1+λ2. Then dimVλ = m+1
and Ωλ is a ”tridiagonal” matrix: all entries [Ωλ]ij of the matrix Ωλ are zeros, except
[Ωλ]k,k = (λ1 − k + 1)E11 + (λ2 + k − 1)E22, k = 1, . . . , m + 1,
[Ωλ]k,k+1 = (m + 1 − k)E21, k = 1, . . . , m,
[Ωλ]k+1,k = kE12, k = 1, . . . , m.
Proposition 3.1. a) Polynomial Dλ(u) is central.
b) Let µ = (µ1 ≥ µ2) be another dominant weight of gln(C). The image of Dλ(u) under
Harish - Chandra isomorphism χ is the following function of µ:
χ(Dλ(u)) =m∏
k=0
(u + (λ1 − k)µ1 + (λ2 + k)µ2 − k). (4)
Proof. a) Let X = X(a, b, c) be a matrix of size (m + 1) × (m + 1) of the form
am bm 0 . . . 0 0 0
cm am−1 bm−1 . . . 0 0 0
. . . . . . . . . . . . . . . . . . . . .
0 0 0 . . . c2 a1 b1
0 0 0 . . . 0 c1 a0
,
where ai, bj , ck are elements of some (noncommutative) algebra. Define detX as in Section
2. Using the principle k-minors of X, the determinant of X can be computed by recursion
178
formula. Denote by Xk the matrix obtained from X by deleting the first (m − k + 1) rows
and the first (m− k + 1) columns, and by I(k) the determinant of Xk. Then detX = I(m+1),
and we have the following recursion:
Lemma 3.2. The determinants I(k) satisfy the recursion realtion
I(k+1) = akI(k) − ckbk I(k−1), (k = 2, ..., n − 1) (5)
with initial conditions I(1) = a0, I(2) = a1a0 − c1b1.
Lemma 3.3. If X(a, b, c) is a tri-diagonal matrix with coefficients ai, bj , ck and X(a′, b′, c′)
is another tri-diagonal matrix with coefficients a′i, b′
j, c′
k with the property
ai = a′i, cj bj = c′
jb′j
for all i = 0, ..., n j = 1, ..., n, then det X(a, b, c) = det X(a′, b′, c′).
Proof. Follows from the recursion relation.
We apply these observations to compute the determinant of the matrix X(a, b, c) =
Ωλ(u) − L.
The following obvious lemma allows to reduce the determinant of non-commutative matrix
(Ωλ(u) − L) to a determinant of a matrix with commutative coefficients.
Lemma 3.4. The subalgebra of U(gl2(C)) generated by E11, E22, (E12 E21) is commutative.
Put h = E11 − E22, a = E12E21. Due to Lemma 3.3 the tridiagonal matrix X(a′, b′, c′)
with coefficients
a′k
= ak = λ1E11 + λ2E22 + u − m + (k − m)(h − 1), k = 0, . . . , m,
b′k
= k a, c′k
= (m − k + 1), k = 1, . . . , m,
has the same determinant as (Ωλ(u) − L). By Lemma 3.4, X(a′, b′, c′) has commutative
coefficients. Hence det (Ωλ(u)−L) equals det (λ1E11 +λ2E22 +u−m+Am) where Am is the
following matrix:
Am =
0 ma 0 . . . 0 0
1 −(h − 1) (m − 1)a . . . 0 0
. . . . . . . . . . . . . . . . . .
0 0 0 . . . (1 − m)(h − 1) a
0 0 0 . . . m −m(h − 1)
with h and a as above.
Lemma 3.5.
det Am =
m∏
k=0
(−m(h − 1)
2+
(m − 2k)
2
√
(h − 1)2 + 4a
)
.
179
Proof. By Lemma 3.3 det Am = (h − 1)m+1det A′m
with
A′m
=
0 ms 0 . . . 0 0
1(s − 1) −1 (m − 1)s . . . 0 0
. . . . . . . . . . . . . . . . . .
0 0 0 . . . 1 − m s
0 0 0 . . . m(s − 1) −m
,
and s is such that s(s − 1) = a/(h − 1)2. This reduces to
s =1 ±
√
1 + 4a/(h − 1)2
2. (6)
The determinant of A′m
is a variant of Sylvester determinant ([1], [4]). It equals
det A′m
=
m∏
k=0
((m − 2k)s − m + k).
With s as in (6) we have:
(h − 1)((m − 2k)s − m + k) =−m(h − 1)
2±
(m − 2k)
2
√
((h − 1)2 + 4a)
and lemma follows. Note that both values of s give the same value of det Am.
We obtain from calculations above
det (Ωλ(u)−L) =
m∏
k=0
(
u +d
2(E11 + E22) −
m
2+
(m − 2k)
2
((E11 − E22 − 1)2 + 4E12E21
) 1
2
)
.
(7)
Observe that((E11 − E22 − 1)2 + 4E12E21
) 1
2 =((∆1 − 1)2 − 4∆2
) 1
2 , and finally we get
Dλ(u) =
m∏
k=0
(
u +d∆1
2−
m
2+
(m − 2k)
2
((∆1 − 1)2 − 4∆2
) 1
2
)
. (8)
The quantity in (8) has coefficients in Wτ -extension of Z(gl2(C)), where Wτ is the translated
Weyl group. But it is easy to see that after expanding the product, we get a polynomial in u
with coefficients in Z(gln(C). We proved the first part of the proposition.
b) The images of the generators of Z(gln(C)) under Harish-Chandra homomorphism χ
are
χ(∆1) = µ1 + µ2, χ(∆2) = µ1(µ2 − 1).
This together with (8) implies (4).
180
4 Characteristic polynomial
Proposition 4.1. For any dominant weight λ of gln(C) there exists a polynomial
Pλ(u) =m∑
k=0
zkuk
with coefficients zk ∈ Z(gln(C)), such that Pλ(−Ωλ) = 0.
This proposition follows from the similar statements for central elements of semisimple
Lie algebras, proved in [6],[2]. In [2] the Harish-Chandra-image of the polynomial Pλ(u) is
also obtained:
χ(Pλ(u)) =∏
(u + (µ, λi) +1
2(2ρ + λi, λi) −
1
2(2ρ + λ, λ)),
where ρ is the half-sum of positive roots, and the product is taken over all weights λi of Vλ.
Corollary 4.2. In case of gl2(C) the image of the polynomial Pλ(u) under Harish-Chandra
homomorphism is
χ(Pλ(u)) =m∏
k=0
(u + (λ1 − k)µ1 + (λ2 + k)µ2 − k(m + 1 − k)) , (9)
where m = λ1 − λ2.
Comparing (9) with (4), we can see that the polynomials Pλ(u) and Dλ(u) are different.
Recall that for a semi-simple Lie algebra g, the algebra U(g) is a deformation of the sym-
metric algebra S(g). Hence, the central polynomials Pλ(u) and Dλ(u) are the deformations
of a polynomial pλ(u), which has coefficients in the ring of invariants I(g) of the adjoint
action of g on S(g). Since I(g) ≃ C[h∗]W , where W is a Weyl group, and h∗ is a dual to
the Cartan subalgebra h, the polynomial pλ(u) can be represented as a polynomial of u with
coefficients in the ring C[h∗]W . We can extend these observations for the case of gln(C). Let
λ1, . . . , λm+1 be the set of weights of Vλ. Then we can write that
pλ(u) =∏
λi
(u + (µ, λi)),
as a (symmetric) function of µ ∈ h∗.
Let us summarize the facts about Dλ(u) and Pλ(u):
a) Dλ(u) and Pλ(u) are deformations of pλ(u), and the case of gl2(C) shows that in general
these are different deformations.
b) In case of vector representation λ = (1), D(1)(u) = P(1)(u) for any gln(C). This is
proved, for example, in [7].
c) Pλ(u) is known to be a central polynomial for all λ for any gln(C).
d) The centrality of Dλ(u) is proved in two cases: for all λ for gl2(C), and for λ = (1) for
gln(C). However, we state the following two conjectures.
181
Conjecture 4.3. For any dominant weight λ there exists a basis of the vector space Vλ such
that the polynomial Dλ(u) has coefficients in the center Z(gln(C)).
Conjecture 4.4. The Harish-Chandra image of Dλ(u) is given by polynomial
∏
(u + (µ + ρ, λi) −m
2),
where the product is taken over all weights λi of the representation Vλ, and dimVλ = m+1.
5 Capelli elements
In this section we show that the shifted determinant can be viewed as some sort of plethysm
applied to Capelli elements.
Consider an element S of U(g) ⊗ EndV , defined by S =∑
ijEij ⊗ Eij. Using the abbre-
viated notation S = S(1) ⊗ S(2), put
S1j = S(1) ⊗ 1⊗(j−1) ⊗ S(2) ⊗ 1⊗(m−j)
(viewed as an element of U(g) ⊗ End(Cn)⊗m).
Let c1, . . . , cM be the set of contents of the standard Young tableau of shape λ, and
let Fλ be a Young symmetrizer that corresponds to the diagram λ:
Fλ : (Cn)⊗M → Vλ, Vλ ⊂ (Cn)⊗M .
Following [11], define an element Sλ(u) of U(gln(C)) ⊗ EndVλ by
Sλ(u) = ((S12 − u − c1) . . . (S1 M+1 − u − cM )) (id ⊗ Fλ). (10)
Then
cλ(u) = tr (Sλ(u))
is the Capelli polynomial, associated to λ. It has coefficients in Z(gln(C)). The theory of
Capelli elements is developed in the papers, mentioned in the Introduction.
Now let π0 be a vector representation, let Sλ(u) = (π0 ⊗ id)Sλ(u), and let Asyms be the
antisymmetrizer, defined as in Section 2.
Proposition 5.1. For λ ⊢ M , dimVλ = (m + 1),
Dλ(u) = f(u) tr
(
Sλ(m − u)12Sλ(m − 1 − u)13 . . . Sλ(−u)1m+2 · Asymm+1
)
, (11)
where
f(u) =
m∏
s=0
M∏
k=1
(u − s)
u − s − ck
.
182
Remark. The formula (11) tells that in order to obtain a shifted determinant for dominant
weight λ, one has to do the following:
1) Take (m + 1) copies of elements Sλ(u).
2) Shift the variable u in each copy by (m − s), s = 0, ...,m.
3) Apply antisymmetrizer of order m + 1 to these shifted (m + 1) copies.
4) Multiply by certain rational function and take the trace.
The result is a shifted analogue of plethysm of two Schur functors: Fλ and F(1m+1) (com-
pare with the definition of Sλ (10)).
Proof. Let (π0, V0 = Cn) be the vector representation of gln(C). Using the Young symmetrizer
Fλ, we construct the element (1 ⊗ Fλ) of End (V M
0), which is a projector from V ⊗M
0to V0⊗Vλ.
We use Proposition 2.12 from [10]:
Proposition 5.2. Let P =∑
Eij ⊗Eji be a permutation matrix on Cn⊗C
n. Let Pk,l denote
the action of this operator on the kth and lth components of the tensor product (Cn)⊗M . Then
M∏
k=1
(
1 −P1, k+1
u − ck
)
(1 ⊗ Fλ) =
(
1 −M∑
k=1
P1, k+1
u
)
(1 ⊗ Fλ), (12)
where Fλ is the Young symmertrizer, and c1, . . . cn are the contents of the standard tableau
of shape λ.
With the standard coproduct δ in U(gln(C)) we obtain:
Ωλ =∑
ij
π0(Eij) ⊗ πλ(Eji) =
∑
ij
Eij ⊗ δ(M)(Eji)
(1 ⊗ Fλ) =
∑
l=1...M,
P1,l+1
(1 ⊗ Fλ),
This implies
Ωλ(u) = u
M∏
k=1
(
1 +P1, k+1
u + ck
)
(1 ⊗ Fλ) ∈ EndV0 ⊗ EndVλ. (13)
Let φ : gln(C) → gln(C) be an automorphism, defined by φ(X) = −X⊤. Put πλ⋆ = πλ φ.
Observe that (φ ⊗ πλ)Ω = id ⊗ (πλ φ)Ω, so
Ωλ⋆(u) = u
M∏
k=1
(
1 −S1, k+1
u + ck
)
(1 ⊗ Fλ) = u
M∏
k=1
(−1)
(u + ck)(π0 ⊗ Id)Sλ(u). (14)
In other words, Ωλ⋆(u) is proportional to the image of Sλ(u) under the map (π0 ⊗ Id) :
U(gln(C)) ⊗ End Vλ → EndV0 ⊗ EndVλ.
The representation X → −(πλ⋆(X))⊤, X ∈ gln(C) is isomorphic to πλ. Thus we can write
in some basis
Ω⊤λ(u) = −Ωλ⋆(−u). (15)
183
Recall from Section 2 that
Dλ(u) = α (Ωλ(u − m), . . . , Ωλ(u)) = tr(Asymm+1 Ωλ(u − m)12 . . . Ωλ(u)1m+2).
It is easy to see that
tr(AsymsA1 ⊗ · · · ⊗ As) = tr(A⊤1 ⊗ · · · ⊗ A⊤
sAsyms)
Hence,
Dλ = tr
(
Ω⊤λ(u − m)1 2 . . . Ω⊤
λ(u)1 m+2 Asymm+1
)
= (−1)m+1 tr ( (Ωλ⋆(−u + m)12 . . . Ωλ⋆(−u)1 m+2)Asymm+1 )
= tr
(m∏
s=0
(u − s)
M∏
k=1
(S1,sM+k+1(u + s − m − ck)
u − s − ck
)
(π0 ⊗ Fλ)⊗(m+1) · Asymm+1
)
Comparing the last formula with the definition of Sλ(u), we obtain (11)
References
[1] R.Askey, Evaluation of some determinants, Proceedings of the 4th ISAAC Congress,
(H. G. W. Begehr et al, eds.), World Scientific (2005) 1–16.
[2] M. D.Gould, Characteristic identities for semisimple Lie algebras, J. Austal. Math. Soc.
B 26 (3) (1985) 257–283.
[3] M. D.Gould, On the matrix elements of the U(n) generators, J. Math. Phys. 22 (1)
(1981) 15–22.
[4] O.Holtz, Evaluation of Sylvester type determinants using block-triangularization Pro-
ceedings of the 4th ISAAC Congress, (H. G. W. Begehr et al, eds.) World Scientific,
(2005) 395–405.
[5] F. Knop, Symmetric and Non-Symmetric Quantum Capelli Polynomials, Com-
ment. Math. Helv. 72 (1) (1997) 84–100.
[6] B. Kostant, On the tensor product of finite and an infinite dimensional representation,
J. Funct. Anal. 20 (4) (1975) 257–285.
[7] A.Molev, Sklyanin determinant, Laplace operators and characteristic identities for clas-
sical Lie algebras, J.Math.Phys. 36 (2) (1995) 923–943.
[8] A.Molev, M. Nazarov, Capelli Identities for Classical Lie algebras, Math.Ann. 313 (2)
(1999) 315–357.
184
[9] A.Molev, M. Nazrov, G. Olshankii, Yangians and Classical Lie algebras, Russian Math.
Surveys 51(2) (1996) 205–282.
[10] M. Nazarov, Yangians and Capelli Identities, ”Kirillov’s Seminar on Representation The-
ory”, Amer. Math. Soc. Translat. 181 (2) (1998) 139–163.
[11] A.Okounkov, Quantum Immants and higher Capelli identities, Transformation Groups
1 (1996) 99–126.
[12] A.Okounkov, G. Olshanski, Shifted Schur functions, St. Petersburg Math. J. 9 (1998)
no.2 239–300.
[13] S. Sahi, The specturm of certain invariant differential operators associated to Hermitian
symmetric spaces, in: ”Lie theory and geometry”, Progress Math 123 Boston Birkha
user 1194, 569–576.
185
186
Graded-simple Lie algebras of type B2 and Jordan
systems covered by a triangle
Erhard Neher∗ Maribel Tocon†
Abstract
We announce a classification of graded-simple Jordan systems covered by a compat-
ible triangle, under some natural assumptions on the abelian group, in order to get the
corresponding classification of graded-simple Lie algebras of type B2.
Keywords: Root-graded Lie algebra, Jordan system, idempotent.
1 Introduction
Graded-simple Lie algebras which also have second compatible grading by a root system
appear in the structure theory of extended affine Lie algebras, which generalize affine Lie
algebras and toroidal Lie algebras. If the root system in question is 3-graded, these Lie
algebras are Tits-Kantor-Koecher algebras of Jordan pairs covered by a grid.
In this note we will consider the case of the root system B2. A centreless B2-graded Lie
algebra is the Tits-Kantor-Koecher algebra of a Jordan pair covered by a triangle. Such a Lie
algebra is graded-simple with respect to a compatible Λ-grading if and only if the Jordan pair
is graded-simple with respect to a Λ-grading which is compatible with the covering triangle
[8]. In [10] we give a classification of graded-simple Jordan systems covered by a triangle that
is compatible with the grading, under some natural assumptions on the abelian group, as
well as the corresponding classification of graded-simple Lie algebras of type B2. Our work
generalizes earlier results of Allison-Gao [1] and Benkart-Yoshii [2], and is an extension of the
structure theory of simple Jordan pairs and Jordan triple systems covered by a triangle due
to McCrimmon-Neher [7].
∗Department of Mathematics and Statistics, University of Ottawa, 585 King Edward Ave., Ottawa, Ontario
K1N 6N5, Canada. E-mail: [email protected]. Supported in part by Discovery Grant #8836 of the Natural
Sciences and Research Council of Canada.†Departamento de Estadıstica, Econometrıa, Inv. Operativa y Org. Empresa, Facultad de Derecho, Puerta
Nueva, 14071, Cordoba, Spain. E-mail: [email protected]. Support from Discovery Grant #8836 of the NSRC
of Canada, MEC and fondos FEDER (Becas posdoctorales and MTM2004-03845), and Junta de Andalucıa
(FQM264) is gratefully acknowledged. This paper was written during the second author’s stay at Department
of Mathematics and Statistics, University of Ottawa, Canada.
187
The aim of this note is to provide an outline of our results for Jordan systems. The details
of proofs will appear in [10]. For unexplained notation we refer the reader to [3] and [4].
2 Graded-simple Lie algebras of type B2
The motivation for our research is two-fold. On the one hand, we would like to advance the
theory of graded Jordan structures, and on the other hand we are interested in certain types
of root-graded Lie algebras. In this section we will describe the second part of our motivation
and how it is related to the first.
Let R be a reduced root system. In the following only the case R = B2 will be of interest,
but the definition below works for any finite, even locally finite reduced root system. We will
assume that 0 ∈ R and denote by Q(R) = Z[R] the root lattice of R. We will consider Lie
algebras defined over a ring of scalars k containing 1
2and 1
3. Let Λ be an arbitrary abelian
group.
Definition 2.1. ([9]) A Lie algebra L over k is called (R,Λ)-graded if
(1) L has a compatible Q(R)- and Λ-gradings,
L = ⊕λ∈ΛLλ and L = ⊕α∈Q(R)Lα,
i.e., using the notation Lλ
α= Lλ ∩ Lα we have
Lα = ⊕λ∈ΛLλ
α, Lλ = ⊕α∈Q(R)L
λ
α, and [Lλ
α, Lκ
β] ⊆ Lλ+κ
α+β,
for λ, κ ∈ Λ, α, β ∈ Q(R),
(2) α ∈ Q(R) : Lα 6= 0 ⊆ R,
(3) L0 =∑
06=α∈R[Lα, L−α], and
(4) for every 0 6= α ∈ R the homogeneous space L0α
contains an element eα that is invertible,
i.e., there exists f−α ∈ L0
−αsuch that hα := [eα, f−α] acts on Lβ, β ∈ R, by
[hα, xβ ] = 〈β, α∨〉xβ, xβ ∈ Lβ.
In particular, (eα, hα, fα) is an sl2-triple.
An (R,Λ)-graded Lie algebra is said to be graded-simple if it does not contain proper
nontrivial Λ-graded ideals and graded-division if every nonzero element in Lλ
α, α 6= 0, is
invertible.
Let now L be a centerless (B2, 0)-graded Lie algebra. It then follows from [8] that L is
the Tits-Kantor-Koecher algebra of a Jordan pair V covered by a triangle: i.e.,
V = V1 ⊕ M ⊕ V2,
188
where Vi = V2(ei), i = 1, 2, and M = V1(e1) ∩ V1(e2), for a triangle (u; e1, e2). Recall that a
triple (u; e1, e2) of nonzero idempotents of V is a triangle if
ei ∈ V0(ej), i 6= j, ei ∈ V2(u), i = 1, 2, and u ∈ V1(e1) ∩ V1(e2),
and the following multiplication rules hold for σ = ±:
Q(uσ)e−σ
i= eσ
j, i 6= j, and Q(eσ
1 , eσ
2 )u−σ = uσ.
If moreover, L is (B2,Λ)-graded, then V is also Λ-graded, i.e., as k-module V σ =⊕
λ∈ΛV σ[λ], σ =
±, with
Q(V σ[λ])V −σ[µ] ⊆ V σ[2λ + µ] and V σ[λ], V −σ[µ], V σ[ν] ⊆ V σ[λ + µ + ν]
for all λ, µ, ν ∈ Λ, σ = ±. A Jordan pair that is Λ-graded and covered by a triangle which
lies in the homogeneous 0-space V [0] is called Λ-triangulated. It therefore follows from the
above that if L is a centerless (B2,Λ)-graded Lie algebra, then L is the Tits-Kantor-Koecher
algebra of a Λ-triangulated Jordan pair V . Moreover, L is graded-simple if and only if V is
graded-simple.
Therefore, one can get a description of graded-simple (B2,Λ)-graded Lie algebras from
the corresponding classification of graded-simple Λ-triangulated Jordan pairs. However, the
classification of graded-simple Λ-triangulated Jordan pairs is only known for Λ = 0 [7].
In what follows, we extend this classification to more general Λ. In doing so, we work with
Jordan structures over arbitrary rings of scalars k. This generality is of independent interest
from the point of view of Jordan theory. Moreover, the simplifications that would arise from
assuming 1
2and 1
3∈ k are minimal, e.g., we could avoid working with ample subspaces in our
two basic examples 3.1 and 3.2.
3 Graded-simple triangulated Jordan triple systems
Let k be an arbitrary ring of scalars and let J be a Jordan triple system over k. Recall that
a triple of nonzero tripotents (u; e1, e2) is called a triangle if ei ∈ J0(ej), i 6= j, ei ∈ J2(u),
i = 1, 2, u ∈ J1(e1) ∩ J1(e2), and the following multiplication rules hold: P (u)ei = ej , i 6= j,
and P (e1, e2)u = u. In this case, e := e1 + e2 is a tripotent such that e and u have the same
Peirce spaces. A Jordan triple system with a triangle (u; e1, e2) is said to be triangulated if
J = J2(e1)⊕(J1(e1)∩ J1(e2)
)⊕ J2(e2) which is equivalent to J = J2(e). In this case, we will
use the notation Ji = J2(ei) and M = J1(e1) ∩ J1(e2). Hence
J = J1 ⊕ M ⊕ J2.
Note that ∗ := P (e)P (u) = P (u)P (e) is an automorphism of J of period 2 such that u∗ = u,
e∗i
= ej , and so J∗i
= Jj.
189
Let Λ be an abelian group. We say that J is Λ-graded if the underlying module is Λ-graded,
say J =⊕
λ∈ΛJλ, and the family (Jλ : λ ∈ Λ) of k-submodules satisfies P (Jλ)Jµ ⊆ J2λ+µ
and Jλ, Jµ, Jν ⊆ Jλ+µ+ν for all λ, µ, ν ∈ Λ. We call J Λ-triangulated if it is Λ-graded and
triangulated by (u; e1, e2) ⊆ J0 and faithfully Λ-triangulated if any x1 ∈ J1 with x1 · u = 0
vanishes, where the product · is defined as follows:
Ji × M → M : (xi,m) 7→ xi · m = L(xi)m := xi, ei,m.
There are two basic models for Λ-triangulated Jordan triple systems:
Example 3.1. Λ-triangulated hermitian matrix systems H2(A,A0, π,−). A Λ-graded (as-
sociative) coordinate system (A,A0, π,−) consists of a unital associative Λ-graded k-algebra
A =⊕
λ∈ΛAλ, a graded submodule A0 =
⊕
λ∈ΛAλ
0for Aλ
0= A0 ∩ Aλ, an involution π and
an automorphism − of period 2 of A. These data satisfy the following conditions: π and −
commute and are both of degree 0, i.e., (Aλ)π = Aλ = Aλ for all λ ∈ Λ, A0 is −-stable and
π-ample in the sense that A0 = A0 ⊆ H(A,π), 1 ∈ A0 and aa0aπ ⊆ A0 for all a ∈ A and
a0 ∈ A0.
To a Λ-graded coordinate system (A,A0, π,− ) we associate the Λ-triangulated hermitian
matrix system H = H2(A,A0, π,− ) which, by definition, is the Jordan triple system of 2× 2-
matrices over A which are hermitian (X = Xπt) and have diagonal entries in A0, with triple
product P (X)Y = XYπt
X = XY X. This system is clearly Λ-graded: H =⊕
λ∈ΛHλ, where
Hλ = spanaλ
0Eii, a
λE12 + (aλ)πE21 : aλ
0∈ Aλ
0, aλ ∈ Aλ, i = 1, 2, and Λ-triangulated by
(u = E12 + E21; e1 = E11, e2 = E22) ⊆ H0.
One can prove that H2(A,A0, π,− ) is graded-simple if and only if (A,π,− ) is graded-
simple. In this case, H2(A,A0, π,− ) is graded isomorphic to one of the following:
(I) H2(A,A0, π,− ) for a graded-simple associative unital A;
(II) Mat2(B) for a graded-simple associative unital B with graded automorphism −, where
(bij) = (bij) for (bij) ∈ Mat2(B) and P (x)y = xyx;
(III) Mat2(B) for a graded-simple associative unital B with graded involution ι, where (bij) =
(bι
ij) for (bij) ∈ Mat2(B) and P (x)y = xytx;
(IV) polarized H2(B,B0, π) ⊕ H2(B,B0, π) for a graded-simple B with graded involution π;
(V) polarized Mat2(B)⊕Mat2(B) for a graded-simple associative unital B and P (x)y = xyx.
The examples (IV) and (V) are special cases of polarized Jordan triple systems. Recall
that a Jordan triple system T is called polarized if there exist submodules T± such that
T = T+ ⊕ T− and for σ = ± we have P (T σ)T σ = 0 = T σ, T σ, T−σ and P (T σ)T−σ ⊆ T σ.
In this case, V = (T+, T−) is a Jordan pair. Conversely, to any Jordan pair V = (V +, V −)
190
we can associate a polarized Jordan triple system T (V ) = V + ⊕ V − with quadratic map P
defined by P (x)y = Q(x+)y− ⊕ Q(x−)y+ for x = x+ ⊕ x− and y = y+ ⊕ y−. In particular,
for any Jordan triple system T the pair (T, T ) is a Jordan pair and hence it has an associated
polarized Jordan triple system which we denote T ⊕T . It is clear that if T is a Λ-triangulated
Jordan triple system then so is T ⊕ T .
Example 3.2. Λ-triangulated ample Clifford systems AC(q, S,D0). This example is a sub-
triple of a full Clifford system which we will define first. It is given in terms of
(i) a Λ-graded unital commutative associative k-algebra D =⊕
λ∈ΛDλ endowed with an
involution − of degree 0,
(ii) a Λ-graded D-module M =⊕
λ∈ΛMλ,
(iii) a Λ-graded D-quadratic form q : M → D, hence q(Mλ) ⊂ D2λ and q(Mλ,Mµ) ⊂ Mλ+µ,
(iv) a hermitian isometry S : M → M of q of order 2 and degree 0, i.e., S(dx) = dS(x) for
d ∈ D, q(S(x)
)= q(x), S2 = Id and S(Mλ) = Mλ, and
(v) u ∈ M0 with q(u) = 1 and S(u) = u.
Given these data, we define
V := De1 ⊕ M ⊕ De2,
where De1 ⊕ De2 is a free Λ-graded D-module with basis (e1, e2) of degree 0. Then V is a
Jordan triple system, called a full Clifford system and denoted by FC(q, S), with respect to
the product
P (c1e1 ⊕ m ⊕ c2e2) (b1e1 ⊕ n ⊕ b2e2) = d1e1 ⊕ p ⊕ d2e2, where
di = c2
ibi + ci q
(m,S(n)
)+ bj q(m)
p = [c1b1 + c2b2 + q(m,S(n)
)]m + [c1c2 − q(m)]S(n).
It is easily seen that FC(q, S) is Λ-triangulated by (u; e1, e2).
But in general we need not take the full Peirce spaces Dei in order to get a Λ-triangulated
Jordan triple system. Indeed, let us define a Clifford-ample subspace of (D, ¯, q) as above as a
Λ-graded k-submodule D0 =⊕
λ∈Λ(D0∩Dλ) such that D0 = D0, 1 ∈ D0 and D0q(M) ⊆ D0.
Then
AC(q, S,D0) := D0 e1 ⊕ M ⊕ D0 e2,
also denoted AC(q,M,S,D,−,D0) if more precision is helpful, is a Λ-graded subsystem of the
full Clifford system FC(q, S) which is triangulated by (u; e1, e2). It is called a Λ-triangulated
ample Clifford system.
191
One can prove that AC(q, S,D0) is graded-simple if and only if q is graded-nondegenerate
(in the obvious sense) and (D,−) is graded-simple. In this case, either D = F is a graded-
division algebra or D is graded isomorphic to F ⊞ F for a graded-division algebra F with −
the exchange automorphism. In the latter case AC(q, S,D0) = AC(q, S, F0)⊕AC(q, S, F0) is
polarized with a Clifford ample subspace F0 ⊆ F .
It is an important fact that one can find the above two examples of Λ-triangulated Jordan
triple systems inside any faithfully Λ-triangulated Jordan triple system. More precisely, let
J = J1 ⊕ M ⊕ J2 be faithfully Λ-triangulated by (u; e1, e2), put C0 = L(J1) and let C
be the subalgebra of Endk(M) generated by C0. Then C is naturally Λ-graded, c 7→ c =
P (e)cP (e) is an automorphism of C of degree 0 and L(x1) · · ·L(xn) 7→ (L(x1) · · ·L(xn))π =
L(xn) · · ·L(x1) induces a (well-defined) involution of C of degree 0. One can prove that the
Λ-graded subsystem
Jh = J1 ⊕ Cu ⊕ J2
is Λ-triangulated by (u; e1, e2) and graded isomorphic to H2(A,A0, π,− ) under the map
x1 ⊕ cu ⊕ y2 7→
(
L(x1) c
cπ L(y∗2)
)
for A = C|Cu, A0 = C0|Cu. Moreover, J has a Λ-graded subsystem
Jq = K1 ⊕ N ⊕ K2
for appropriately defined submodules Ki ⊂ Ji and N ⊂ M , which is Λ-triangulated by
(u; e1, e2) and graded isomorphic to AC(q, S,D0) under the map
x1 ⊕ n ⊕ x2 7→ L(x1) ⊕ n ⊕ L(x∗2)
where D0 = L(K1), D is the subalgebra of Endk(N) generated by D0, q(n) = L(P (n)e2) and
S(n) = P (e)n. (Roughly speaking, Jq is the biggest graded subsystem of J where the identity
(x1 − x∗1) · N ≡ 0 holds). Moreover, the two isomorphisms above map the triangle (u; e1, e2)
of J onto the standard triangle of H2(A,A0, π,− ) or AC(q, S,D0), respectively. We note that
for Λ = 0 these two partial coordinatization theorems were proven in [7].
A question that arises naturally is the following: Let J be faithfully Λ-triangulated by
(u; e1, e2). When is Jh or Jq the whole J?
(i) If M = Cu, then J = Jh and thus J is graded isomorphic to a hermitian matrix system,
(ii) If u is C-faithful and (x1 − x∗1) · m = 0 for all x1 ∈ J1 and m ∈ M , then J = Jq and
thus J is graded isomorphic to an ample Clifford system.
One can show that (ii) holds whenever C is commutative and −-simple. In fact, (i) or (ii)
above holds if (C, π,−) is graded-simple.
192
Proposition 3.3. Let J be a graded-simple Λ-triangulated Jordan triple system satisfying
one of the following conditions
(a) every m ∈ M is a linear combination of invertible homogeneous elements, or
(b) Λ is torsion-free.
Then (C, π,−) is graded-simple. In this case, u is C-faithful and M = Cu or C is commuta-
tive.
All together, we have the following result:
Theorem 3.4. A graded-simple Λ-triangulated Jordan triple system satisfying (a) or (b) of
Prop. 3.3 is graded isomorphic to one of the following:
a non polarized Jordan triple system
(I) H2(A,A0, π,− ) for a graded-simple A with graded involution π and automorphism −;
(II) Mat2(B) with P (x)y = xyx for a graded-simple associative unital B with graded auto-
morphism − and (yij) = (yij) for (yij) ∈ Mat2(B);
(III) Mat2(B) with P (x)y = xytx for a graded-simple associative unital B with graded invo-
lution ι and (yij) = (yι
ij) for (yij) ∈ Mat2(B);
(IV) AC(q, S, F0) for a graded-nondegenerate q over a graded-division F with Clifford-ample
subspace F0;
or a polarized Jordan triple system
(V) H2(B,B0, π) ⊕ H2(B,B0, π) for a graded-simple B with graded involution π;
(VI) Mat2(B) ⊕ Mat2(B) for a graded-simple associative unital B with P (x)y = xyx;
(VII) AC(q, S, F0) ⊕ AC(q, S, F0) for AC(q, S, F0) as in (IV).
Conversely, all Jordan triple systems in (I)-(VII) are graded-simple Λ-triangulated.
Since Λ = 0 is a special case of our assumption (b), the theorem above generalizes [7,
Prop. 4.4].
4 Graded-simple triangulated Jordan pairs and algebras
We consider Jordan algebras and Jordan pairs over arbitrary rings of scalars. In order to apply
our results, we will view Jordan algebras as Jordan triple systems with identity elements.
Thus, to a Jordan algebra J we associate the Jordan triple system T (J) defined on the k-
module J with Jordan triple product Pxy = Uxy. The element 1J ∈ J satisfies P (1J ) = Id.
193
Conversely, every Jordan triple system T containing an element 1 ∈ T with P (1) = Id is
a Jordan algebra with unit element 1 and multiplication Uxy = Pxy. A Λ-graded Jordan
algebra J is called Λ-triangulated by (u; e1, e2) if ei = e2
i∈ J0, i = 1, 2, are supplementary
orthogonal idempotents and u ∈ J1(e1)0 ∩ J1(e2)
0 with u2 = 1 and u3 = u. Thus, with
our definition of a triangle in a Jordan algebra, J is Λ-triangulated by (u; e1, e2) iff T (J) is
Λ-triangulated by (u; e1, e2).
This close relation to Λ-triangulated Jordan triple systems also indicates how to get exam-
ples of Λ-triangulated Jordan algebras: We take a Jordan triple system which is Λ-triangulated
by (u; e1, e2) and require P (e) = Id for e = e1 + e2. Doing this for our two basic examples 3.1
and 3.2, yields the following examples of Λ-triangulated Jordan algebras.
(A) Hermitian matrix algebra: This is the Jordan triple system H2(A,A0, π,− ) with − =
Id, which we will write as H2(A,A0, π). Note that this is a Jordan algebra with product
U(x)y = P (x)y = xyx and identity element E = E11 + E22. If, for example, A = B ⊞ Bop
with π the exchange involution, then H2(A,A0, π) is graded isomorphic to Mat2(B) where
Mat2(B) is the Jordan algebra with product Uxy = xyx.
(B) Quadratic form Jordan algebra: This is the ample Clifford system AC(q, S,D,−,D0)
with − = Id and S|M = Id. Since then P (e) = Id we get indeed a Λ-triangulated Jordan
algebra denoted ACalg(q,D,D0). Note that this Jordan algebra is defined on D0e1⊕M⊕D0e2
and has product Uxy = q(x, y)x − q(x)y where q(d1e1 ⊕ m ⊕ d2e2) = d1d2 − q(m) and
(d1e1 ⊕ m ⊕ d2e2) = d2e1 ⊕ −m ⊕ d1e1. (If 1
2∈ k it is therefore a reduced spin factor in the
sense of [6, II, §3.4].)From the classification given in Th. 3.4 we get:
Theorem 4.1. A graded-simple Λ-triangulated Jordan algebra satisfying
(a) every m ∈ M is a linear combination of invertible homogeneous elements of M , or
(b) Λ is torsion-free,
is graded isomorphic to one of the following Jordan algebras:
(I) H2(A,A0, π) for a graded-simple A with graded involution π;
(II) Mat2(B) for a graded-simple associative unital B;
(III) ACalg(q, F, F0) for a graded-nondegenerate q : M → F over a commutative graded-
division algebra F and a Clifford-ample subspace F0.
Conversely, all Jordan algebras in (I)–(III) are graded-simple Λ-triangulated.
Note that for Λ = 0 this theorem generalizes the well known Capacity Two Theorem
for Jordan algebras.
194
With the above algebra classification at hand and taking into account that a Λ-triangulated
Jordan pair can be viewed as a disguised Λ-triangulated Jordan algebra, we get the following
classification of Λ-triangulated Jordan pairs:
Theorem 4.2. A graded-simple Λ-triangulated Jordan pair satisfying
(a) every m ∈ Mσ is a linear combination of invertible homogeneous elements of Mσ, or
(b) Λ is torsion-free,
is graded isomorphic to a Jordan pair (J, J) where
(I) J = H2(A,A0, π) is the hermitian matrix algebra of a graded-simple A with graded
involution π;
(II) J = Mat2(B) for a graded-simple associative unital B;
(III) J = AC(q, Id, F0) for a graded-nondegenerate q over a graded-division algebra F with
Clifford-ample subspace F0.
Conversely, all Jordan pairs described above are graded-simple Λ-triangulated.
Note that a Jordan pair V satisfies assumption (a) if it is the Jordan pair associated to a
graded-division (B2,Λ)-graded Lie algebra for an arbitrary Λ (cf. §1).References
[1] B. N. Allison and Y. Gao, The root system and the core of an extended affine Lie algebra,
Selecta math. New ser., 7, (2001), 149-212.
[2] G. Benkart and Y. Yoshii, Lie G-tori of symplectic type, to appear in Quarterly J. Math.,
(2006).
[3] N. Jacobson, Structure theory of Jordan algebras, University of Arkansas Lecture Notes
in Mathematics 5, University of Arkansas, Fayetteville, Ark., (1981), 317 pp.
[4] O. Loos, Jordan Pairs, Lectures Notes in Mathematics, Vol. 460, Springer-Verlag, New
York, (1975).
[5] K. McCrimmon, Peirce ideals in Jordan triple systems, Pacific J. Math, 83, 2 (1979),
415-439.
[6] K. McCrimmon, A taste of Jordan algebras, Universitext. Springer-Verlag, New York,
(2004). xxvi+562 pp.
[7] K. McCrimmon and E. Neher, Coordinatization of Triangulated Jordan Systems, J.
Algebra, 114, 2 (1998), 411-451.
195
[8] E. Neher, Lie algebras graded by 3-graded root systems and Jordan pairs covered by grids,
Amer. J. Math., 118, (1996), 439-491.
[9] E. Neher, Extended affine Lie algebras and Lie tori, Lecture Notes, Spring 2006.
[10] E. Neher and M. Tocon, Graded-simple Lie algebras of type B2 and graded-simple Jordan
pairs covered by a triangle, preprint 2007.
196
Restricted simple Lie algebras and their
infinitesimal deformations
Filippo Viviani∗
Abstract
In the first two sections, we review the Block-Wilson-Premet-Strade classification of
restricted simple Lie algebras. In the third section, we compute their infinitesimal de-
formations. In the last section, we indicate some possible generalizations by formulating
some open problems.
Keywords: Restricted simple Lie algebras, Deformations.
1 Restricted Lie algebras
We fix a field F of characteristic p > 0 and we denote with Fp the prime field with p elements.
All the Lie algebras that we will consider are of finite dimension over F . We are interested
in particular class of Lie algebras, called restricted (or p-Lie algebras).
Definition 1.1 (Jacobson [JAC37]). A Lie algebra L over F is said to be restricted (or a
p-Lie algebra) if there exits a map (called p-map), [p] : L → L, x 7→ x[p], which verifies the
following conditions:
1. ad(x[p]) = ad(x)[p] for every x ∈ L.
2. (αx)[p] = αpx[p] for every x ∈ L and every α ∈ F .
3. (x0 + x1)[p] = x
[p]
0+ x
[p]
1+∑
p−1
i=1si(x0, x1) for every x, y ∈ L, where the element
si(x0, x1) ∈ L is defined by
si(x0, x1) = −1
r
∑
u
adxu(1) adxu(2) · · · adxu(p−1)(x1),
the summation being over all the maps u : [1, · · · , p − 1] → 0, 1 taking r-times the
value 0.
∗Universita degli studi di Roma Tor Vergata, Dipartimento di Matematica, via della Ricerca Scientifica 1,
00133 Rome. E-mail: [email protected]. The author was supported by a grant from the Mittag-
Leffler Institute of Stockholm.
197
Example. 1. Let A an associative F -algebra. Then the Lie algebra DerF A of F -derivations
of A is a restricted Lie algebra with respect to the p-map D 7→ Dp := D · · · D.
2. Let G a group scheme over F . Then the Lie algebra Lie(G) associated to G is a restricted
Lie algebra with respect to the p-map given by the differential of the homomorphism
G→ G, x 7→ xp := x · · · x.
One can naturally ask when a F -Lie algebra can acquire the structure of a restricted Lie
algebra and how many such structures there can be. The following criterion of Jacobson
answers to that question.
Proposition 1.2 (Jacobson). Let L be a Lie algebra over F . Then
1. It is possible to define a p-map on L if and only if, for every element x ∈ L, the p-th
iterate of ad(x) is still an inner derivation.
2. Two such p-maps differ by a semilinear map from L to the center Z(L) of L, that is a
map f : L→ Z(L) such that f(αx) = αpf(x) for every x ∈ L and α ∈ F .
Proof. See [JAC62, Chapter V.7].
Many of the modular Lie algebras that arise “in nature” are restricted. As an example of
this principle, we would like to recall the following two results from the theory of finite group
schemes and the theory of inseparable field extensions.
Theorem 1.3. There is a bijective correspondence
Restricted Lie algebras/F ←→ Finite group schemes/F of height 1,
where a finite group scheme G has height 1 if the Frobenius F : G → G(p) is zero. Explicitly
to a finite group scheme G of height 1, one associates the restricted Lie algebra Lie(G) :=
T0G. Conversely, to a restricted Lie algebra (L, [p]), one associates the finite group scheme
corresponding to the dual of the restricted enveloping Hopf algebra U [p](L) := U(L)/(xp−x[p]).
Proof. See [DG70, Chapter 2.7].
Theorem 1.4. Suppose that [F : F p] <∞. There is a bijective correspondence
Inseparable subextensions of exponent 1 ←→ Restricted subalgebras of Der(F )
where the inseparable subextensions of exponent 1 are the subfields E ⊂ F such that F p ⊂ E ⊂
F and Der(F ) := DerFp(F ) = DerF p(F ). Explicitly to any field F p ⊂ E ⊂ F one associates
the restricted subalgebra DerE(F ). Conversely, to any restricted subalgebra L ⊂ Der(F ), one
associates the subfield EL := x ∈ F |D(x) = 0 for all D ∈ L.
Proof. See [JAC80, Chapter 8.16].
198
2 Classification of restricted simple Lie algebras
Simple Lie algebras over an algebraically closed field of characteristic zero were classified
at the beginning of the XIX century by Killing and Cartan. The classification proceeds
as follows: first the non-degeneracy of the Killing form is used to establish a correspondence
between simple Lie algebras and irreducible root systems and then the irreducible root systems
are classified by mean of their associated Dynkin diagrams. It turns out that there are four
infinite families of Dynkin diagrams, called An, Bn, Cn, Dn, and five exceptional Dynkin
diagram, called E6, E7, E8, F4 and G2. The four infinite families correspond, respectively, to
the the special linear algebra sl(n+1), the special orthogonal algebra of odd rank so(2n+1),
the symplectic algebra sp(2n) and the special orthogonal algebra of even rank so(2n). For the
simple Lie algebras corresponding to the exceptional Dynkin diagrams, see the book [JAC71]
or the nice account in [BAE02].
These simple Lie algebras admits a model over the integers via the (so-called) Chevalley
bases. Therefore, via reduction modulo a prime p, one obtains a restricted Lie algebra over
Fp, which is simple up to a quotient by a small ideal. For example sl(n) is not simple if p
divide n, but its quotient psl(n) = sl(n)/(In) by the unit matrix In becomes simple. There
are similar phenomena occuring only for p = 2, 3 for the other Lie algebras (see [STR04,
Page 209] or [SEL67]). The restricted simple algebras obtained in this way are called alge-
bras of classical type. Their Killing form is non-degenerate except at a finite number of
primes. Moreover, they can be characterized as those restricted simple Lie algebras admitting
a projective representation with nondegenerate trace form (see [BLO62], [KAP71]).
However, there are restricted simple Lie algebras which have no analogous in charac-
teristic zero and therefore are called nonclassical. The first example of a nonclassical re-
stricted simple Lie algebra is due to E. Witt, who in 1937 realized that the derivation algebra
W (1) := DerF (F [X]/(Xp)) over a field F of characteristic p > 3 is simple with a degener-
ate Killing form. In the succeeding three decades, many more nonclassical restricted simple
Lie algebras have been found (see [JAC43], [FRA54], [AF54], [FRA64]). The first compre-
hensive conceptual approach to constructing these nonclassical restricted simple Lie algebras
was proposed by Kostrikin and Shafarevich in 1966 (see [KS66]). They showed that all the
known examples can be constructed as finite-dimensional analogues of the four classes of
infinite-dimensional complex simple Lie algebras, which occurred in Cartan’s classification
of Lie pseudogroups (see [CAR09]). These restricted simple Lie algebras, called of Cartan-
type, are divided into four families, called Witt-Jacobson, Special, Hamiltonian and Contact
algebras.
Definition 2.1. Let A(n) := F [x1, · · · , xn]/(xp
1, · · · , xp
n) the algebra of p-truncated polyno-
mials in n variables. Then the Witt-Jacobson Lie algebra W (n) is the derivation algebra of
A(n):
W (n) = DerF A(n).
199
For every j ∈ 1, . . . , n, we put Dj := ∂
∂xj. The Witt-Jacobson algebra W (n) is a free
A(n)-module with basis D1, . . . ,Dn. Hence dimF W (n) = npn with a basis over F given by
xaDj | 1 ≤ j ≤ n, xa ∈ A(n).
The other three families are defined as m-th derived algebras of the subalgebras of deriva-
tions fixing a volume form, a Hamiltonian form and a contact form, respectively. More
precisely, consider the natural action of W (n) on the exterior algebra of differential forms in
dx1, · · · ,dxn over A(n). Define the following three forms, called volume form, Hamiltonian
form and contact form:
ωS = dx1 ∧ · · · dxn,
ωH =
m∑
i=1
dxi ∧ dxi+m if n = 2m,
ωK = dx2m+1 +m∑
i=1
(xi+mdxi − xidxi+m) if n = 2m + 1.
Definition 2.2. Consider the following three subalgebras of W (n):
S(n) = D ∈W (n) |DωS = 0,
H(n) = D ∈W (n) |DωH = 0,
K(n) = D ∈W (n) |DωK ∈ A(n)ωK.
Then the Special algebra S(n) (n ≥ 3) is the derived algebra of S(n), while the Hamiltonian
algebra H(n) (n = 2m ≥ 2) and the Contact algebra K(n) (n = 2m + 1 ≥ 3) are the second
derived algebras of H(n) and K(n), respectively.
We want to describe more explicitly the above algebras, starting from the Special algebra
S(n). For every 1 ≤ i, j ≤ n consider the following maps
Dij = −Dji :
A(n) −→W (n)
f 7→ Dj(f)Di −Di(f)Dj .
Proposition 2.3. The algebra S(n) has F -dimension equal to (n−1)(pn−1) and is generated
by the elements Dij(xa) for xa ∈ A(n) and 1 ≤ i < j ≤ n.
Proof. See [FS88, Chapter 4.3].
Suppose now that n = 2m ≥ 2 and consider the map DH : A(n)→W (n) defined by
DH(f) =m∑
i=1
[Di(f)Di+m −Di+m(f)Di] ,
where, as before, Di := ∂
∂xi∈ W (n). Then the Hamiltonian algebra can be described as
follows:
200
Proposition 2.4. The above map DH induces an isomorphism
DH : A(n)6=1,xσ
∼=−→ H(n),
where A(n)6=1,xσ = xa ∈ A(n) | xa 6= 1, xa 6= xσ := xp−1
1· · · xp−1
n . Therefore H(n) has
dimension pn − 2.
Proof. See [FS88, Chapter 4.4].
Suppose finally that n = 2m + 1 ≥ 3. Consider the map DK : A(n)→ K(n) defined by
DK(f) =
m∑
i=1
[Di(f)Di+m −Di+m(f)Di] +
2m∑
j=1
xj [Dn(f)Dj −Dj(f)Dn] + 2fDn.
Then the Contact algebra can be described as follows:
Proposition 2.5. The above map DK induces an isomorphism
K(n) ∼=
A(n) if p 6 | (m + 2),
A(n)6=xτ if p | (m + 2),
where A(n)6=xτ := xa ∈ A(n) | xa 6= xτ := xp−1
1· · · xp−1
n . Therefore K(n) has dimension pn
if p 6 | (m + 2) and pn − 1 if p | (m + 2).
Proof. See [FS88, Chapter 4.5].
Kostrikin and Shafarevich (in the above mentioned paper [KS66]) conjectured that a
restricted simple Lie algebras (that is a restricted algebras without proper ideals) over an
algebraically closed field of characteristic p > 5 is either of classical or Cartan type. The
Kostrikin-Shafarevich conjecture was proved by Block-Wilson (see [BW84] and [BW88])
for p > 7, building upon the work of Kostrikin-Shafarevich ([KS66] and [KS69]), Kac ([KAC70]
and [KAC74]), Wilson ([WIL76]) and Weisfailer ([WEI78]).
Recently, Premet and Strade (see [PS97], [PS99], [PS01], [PS04]) proved the Kostrikin-
Shafarevich conjecture for p = 7. Moreover they showed that for p = 5 there is only one
exception, the Melikian algebra ([MEL80]), whose definition is given below.
Definition 2.6. Let p = char(F ) = 5. Let W (2) be a copy of W (2) and for an element
D ∈ W (2) we indicate with D the corresponding element inside W (2). The Melikian algebra
M is defined as
M = A(2)⊕W (2)⊕ W (2),
with Lie bracket defined by the following rules (for all D,E ∈W (2) and f, g ∈ A(2)):
[D, E] := [D,E] + 2div(D)E,
[D, f ] := D(f)− 2 div(D)f,
[f1D1 + f2D2, g1D1 + g2D2] := f1g2 − f2g1,
[f, E] := fE,
[f, g] := 2 (gD2(f)− fD2(g))D1 + 2 (fD1(g) − gD1(f))D2,
201
where div(f1D1 + f2D2) := D1(f1) + D2(f2) ∈ A(2).
In characteristic p = 2, 3, there are many exceptional restricted simple Lie algebras (see
[STR04, page 209]) and the classification seems still far away.
3 Infinitesimal Deformations
An infinitesimal deformation of a Lie algebra L over a field F is a Lie algebra L′ over F [ǫ]/(ǫ2)
such that L′ ×F [ǫ]/(ǫ2) F ∼= L. Explicitly, L′ = L + ǫL with Lie bracket [−,−]′ defined by (for
any two elements X,Y ∈ L ⊂ L′):
[X,Y ]′ = [X,Y ] + ǫf(X,Y ),
where [−.−] is the Lie bracket of L and f(−,−) is an 2-alternating function from L to L, con-
sidered a module over itself via adjoint representation. The Jacobi identity for [−,−]′ forces
f to be a cocycle and moreover one can check that two cocycles differing by a coboundary
define isomorphic Lie algebras. Therefore the infinitesimal deformations of a Lie algebra L
are parametrized by the second cohomology H2(L,L) of the Lie algebra with values in the
adjoint representation (see [GER64] for a rigorous treatment).
It is a classical result that simple Lie algebras in characteristic zero are rigid. We want to
give a sketch of the proof of the following Theorem (see [HS97] for details).
Theorem 3.1. Let L be a simple Lie algebra over a field F of characteristic 0. Then, for
every i ≥ 0, we have that
H i(L,L) = 0.
Sketch of the Proof. Since the Killing form β(x, y) = tr(ad(x)ad(y)) is non-degenerate (by
Cartan’s criterion), we can choose two bases ei and e′i of L such that β(ei, e
′j) = δij .
Consider the Casimir element C :=∑
iei ⊗ e′
iinside the enveloping algebra U(L). One can
check that:
1. C belongs to the center of the enveloping algebra and therefore it induces an L-
homomorphism C : L → L, where L is a consider a module over itself via adjoint
action. Moreover since trL(C) = dim(L) 6= 0, C is non-zero and hence is an isomor-
phism by the simplicity of L.
2. The map induced by C on the exact complex U(L)⊗k
∧n
Ln → F is homotopic to 0.
Therefore the induced map on cohomology C∗ : H∗(L,L) → H∗(L,L) is an isomorphism by
(1) and the zero map by (2), which implies that H∗(L,L) = 0.
The above proof uses the non-degeneracy of the Killing form and the non-vanishing of the
trace of the Casimir element, which is equal to the dimension of the Lie algebra. Therefore
202
the same proof works also for the restricted simple Lie algebras of classical type over a field
of characteristic not dividing the determinant of the Killing form and the dimension of the
Lie algebra. Actually Rudakov (see [RUD71]) showed that such Lie algebras are rigid if the
characteristic of the base field is greater or equal to 5 while in characteristic 2 and 3 there
are non-rigid classical Lie algebras (see [CHE05], [CK00], [CKK00]).
It was already observed by Kostrikin and Dzumadildaev ([DK78], [DZU80], [DZU81] and
[DZU89]) that Witt-Jacobson Lie algebras admit infinitesimal deformations. More precisely:
in [DK78] the authors compute the infinitesimal deformations of the Jacobson-Witt algebras
of rank 1, while in [DZU80, Theorem 4], [DZU81] and [DZU89] the author describes the
infinitesimal deformations of the Jacobson-Witt algebras of any rank but without a detailed
proof.
In the papers [VIV1], [VIV2] and [VIV3], we computed the infinitesimal deformations
of the restricted simple Lie algebras of Cartan type in characteristic p ≥ 5, showing in
particular that they are non-rigid. Before stating the results, we need to recall the definition
of the Squaring operators ([GER64]). The Squaring of a derivation D : L→ L is the 2-cochain
defined, for any x, y ∈ L as it follows
Sq(D)(x, y) =
p−1∑
i=1
[Di(x),Dp−i(y)]
i!(p − i)!, (3.1)
where Di is the i-iteration of D. Using the Jacobi identity, it is straightforward to check that
Sq(D) is a 2-cocycle and therefore it defines a class in the cohomology group H2(L,L), which
we will continue to call Sq(D) (by abuse of notation). Moreover for an element γ ∈ L, we
define Sq(γ) := Sq(ad(γ)).
Theorem 3.2. We have that
H2(W (n),W (n)) =
n⊕
i=1
〈Sq(Di)〉F .
Theorem 3.3. We have that
H2(S(n), S(n)) =n⊕
i=1
〈Sq(Di)〉F⊕
〈Θ〉F ,
where Θ is defined by Θ(Di,Dj) = Dij(xτ ).
Theorem 3.4. Let n = 2m + 1 ≥ 3. Then we have that
H2(K(n),K(n)) =
2m⊕
i=1
〈Sq(DK(xi))〉F⊕
〈Sq(DK(1))〉F .
Before stating the next theorem, we need some notations about n-tuples of natural num-
bers. We consider the order relation inside Nn given by a = (a1, · · · , an) ≤ b = (b1, · · · , bn)
203
if ai ≤ bi for every i = 1, · · · , n. We define the degree of a ∈ Nn as |a| =
∑n
i=1ai and
the factorial as a! =∏
n
i=1ai!. For two multindex a, b ∈ N
n such that b ≤ a, we set(a
b
):=∏
n
i=1
(ai
bi
)= a!
b!(a−b)!. For every integer j ∈ 1, · · · , n we call ǫj the n-tuple having
1 at the j-th entry and 0 outside. We denote with σ the multindex (p− 1, · · · , p− 1).
Assuming now that n = 2m, we define the sign σ(j) and the conjugate j′ of 1 ≤ j ≤ 2m
as follows:
σ(j) =
1 if 1 ≤ j ≤ m,
−1 if m < j ≤ 2m,and j′ =
j + m if 1 ≤ j ≤ m,
j −m if m < j ≤ 2m.
Given a multindex a = (a1, · · · , a2m) ∈ N2m, we define the sign of a as σ(a) =
∏σ(i)ai and
the conjugate of a as the multindex a such that ai = ai′ for every 1 ≤ i ≤ 2m.
Theorem 3.5. Let n = 2m ≥ 2. Then if n ≥ 4 we have that
H2(H(n),H(n)) =n⊕
i=1
〈Sq(DH(xi))〉F⊕
i<j
j 6=i′
〈Πij〉F
m⊕
i=1
〈Πi〉F⊕
〈Φ〉F ,
where the above cocycles are defined (and vanish outside) by
Πij(DH(xa),DH(xb)) = DH(xp−1
i′x
p−1
j′[Di(x
a)Dj(xb)−Di(x
b)Dj(xa)]),
Πi(DH(xixa),DH(xi′x
b)) = DH(xa+b+(p−1)ǫi+(p−1)ǫi′ ),
Πi(DH(xk),DH(xσ−(p−1)ǫi−(p−1)ǫi′ )) = −σ(k)DH(xσ−ǫk′ ) for 1 ≤ k ≤ n,
Φ(DH(xa),DH(xb)) =∑
δ≤a,b
|δ|=3
(a
δ
)(b
δ
)
σ(δ) δ! DH(xa+b−δ−δ).
If n = 2 then we have that
H2(H(2),H(2)) =2⊕
i=1
〈Sq(DH(xi))〉F⊕
〈Φ〉F .
Theorem 3.6. We have that
H2(M,M) = 〈Sq(1)〉F
2⊕
i=1
〈Sq(Di)〉F
2⊕
i=1
Sq(Di)〉F .
4 Open Problems
Simple Lie algebras (not necessarily restricted) over an algebraically closed field F of
characteristic p 6= 2, 3 have been classified by Strade and Wilson for p > 7 (see [SW91],
[STR89], [STR92], [STR91], [STR93], [STR94], [STR98]) and by Premet-Strade for p = 5, 7
(see [PS97], [PS99], [PS01], [PS04]). The classification says that for p ≥ 7 a simple Lie algebra
204
is of classical type (and hence restricted) or of generalized Cartan type. Those latter are
generalizations of the Lie algebras of Cartan type, obtained by considering higher truncations
of divided power algebras (not just p-truncated polynomial algebras) and by considering only
the subalgebra of (the so called) special derivations (see [FS88] or [STR04] for the precise
definitions). Again in characteristic p = 5, the only exception is represented by the generalized
Melikian algebras. Therefore an interesting problem would be the following:
Problem 1. Compute the infinitesimal deformations of the simple Lie algebras.
Note that there is an important distinction between restricted simple Lie algebras and
simple restricted Lie algebras. The former algebras are the restricted Lie algebras which do
not have any nonzero proper ideal, while the second ones are the restricted Lie algebras which
do not have any nonzero proper restricted ideal (or p-ideal), that is an ideal closed under the
p-map. Clearly every restricted simple Lie algebra is a simple restricted Lie algebra, but a
simple restricted Lie algebra need not be a simple Lie algebras. Indeed we have the following
Proposition 4.1. There is a bijection
Simple restricted Lie algebras ←→ Simple Lie algebras.
Explicitly to a simple restricted Lie algebra (L, [p]) we associates its derived algebra [L,L].
Conversely to a simple Lie algebra M we associate the restricted subalgebra M [p] of DerF (M)
generated by ad(M) (which is called the universal p-envelope of M).
Proof. We have to prove that the above maps are well-defined and are inverse one of the
other.
• Consider a simple restricted Lie algebra (L, [p]). The derived subalgebra [L,L] ⊳ L is a
non-zero ideal (since L can not be abelian) and therefore [L,L]p = L, where [L,L]p denotes
the p-closure of [L,L] inside L.
Take a non-zero ideal 0 6= I ⊳ [L,L]. Since [L,L]p = L, we deduce from [FS88, Chapter 2,
Prop. 1.3] that I is also an ideal of L and therefore Ip = L by restricted simplicity of (L, [p]).
From loc. cit., it follows also that [L,L] = [Ip, Ip] = [I, I] ⊂ I from which we deduce that
I = L. Therefore [L,L] is simple.
Since ad : L → DerF (L) is injective and [L,L]p = L, it follows by loc. cit. that ad :
L → DerF ([L,L]) is injective. Therefore we have that [L,L] ⊂ L ⊂ DerF ([L,L]) and hence
[L,L][p] = [L,L]p = L.
• Conversely, start with a simple Lie algebra M and consider its universal p-envelop
M < M [p] < DerF (M).
Take any restricted ideal I ⊳p M [p]. By loc. cit., we deduce [I,M [p]] ⊂ I ∩ [M [p],M [p]] =
I ∩ [M,M ] = I ∩M ⊳M . Therefore, by the simplicity of M , either I ∩M = M or I ∩M = 0.
In the first case, we have that M ⊂ I and therefore M [p] = I. in the second case, we have
205
that [I,M [p]] = 0 and therefore I = 0 because M [p] has trivial center. We conclude that M [p]
is simple restricted.
Moreover, by loc. cit., we have that [M [p],M [p]] = [M,M ] = M .
Therefore the preceding classifications of simple Lie algebras (for p 6= 2, 3) give a classifi-
cation of simple restricted Lie algebras.
Problem 2. Compute the infinitesimal deformations of the simple restricted Lie algebras.
There is an important connection between simple restricted Lie algebras and simple
finite group schemes.
Proposition 4.2. Over an algebraically closed field F of characteristic p > 0, a simple finite
group scheme is either the constant group scheme associated to a simple finite group or it is
the finite group scheme of height 1 associated to a simple restricted Lie algebra.
Proof. Let G be a simple finite group scheme. The kernel of the Frobenius map F : G→ G(p)
is a normal subgroup and therefore, by the simplicity of G, we have that either Ker(F ) = 0
or Ker(F ) = G. In the first case, the group G is constant (since F = F ), and therefore it
corresponds to an (abstract) simple finite group. In the second case, the group G is of height
1 and therefore the result follows from Proposition 1.3.
The following problem seems very interesting.
Problem 3. Compute the infinitesimal deformations of the simple finite group schemes.
Since constant finite group schemes (or more generally etale group schemes) are rigid,
one can restrict to the simple finite group schemes of height 1 associated to the simple re-
stricted Lie algebras. Moreover, if (L, [p]) is the simple restricted Lie algebra corresponding
to the simple finite group scheme G, then the infinitesimal deformations of G correspond to
restricted infinitesimal deformations of (L, [p]), that are infinitesimal deformations that admit
a restricted structure. These are parametrized by the second restricted cohomology group
H2∗ (L,L) (defined in [HOC54]). Therefore the above Problem 3 is equivalent to the following:
Problem 4. Compute the restricted infinitesimal deformations of the simple restricted Lie
algebras.
The above Problem 4 is closely related to Problem 2 because of the following spectral
sequence relating the restricted cohomology to the ordinary one (see [FAR91]):
Ep,q
2= HomFrob
(q∧
L,Hp
∗ (L,L)
)
⇒ Hp+q(L,L),
where HomFrob denote the homomorphisms that are semilinear with respect to the Frobenius.
Acknowledgements. The results presented here constitute my PHD thesis. I thank my
supervisor prof. R. Schoof for useful advices and constant encouragement.
206
References
[AF54] A. A. Albert and M. S. Frank: Simple Lie algebras of characteristic p. Rend. Sem.
Mat. Univ. Politec. Torino 14 (1954-1955), 117–139.
[BAE02] J. Baez: The Octonions. Bull. Amer. Math. Soc. 39 (2002), 145-205.
[BLO62] R. E. Block: Trace forms on Lie algebras. Canad. J. Math. 14 (1962), 553–564.
[BW84] R. E. Block and R. L. Wilson: The restricted simple Lie algebras are of classical or
Cartan type. Proc. Nat. Aca. Sci. USA 81 (1984), 5271–5274.
[BW88] R.E. Block and R.L. Wilson: Classification of the restricted simple Lie algebras. J.
Algebra 114 (1988), 115–259.
[CAR09] E. Cartan: Les groupes de transformations continus, infinis, simples. Ann. Sci.
Ecole Nor. Sup. 26 (1909), 93–161.
[CHE05] N. G. Chebochko: Deformations of classical Lie algebras with a homogeneous root
system in characteristic two. I (Russian). Mat. Sb. 196 (2005), no. 9, 125–156.
English translation: Sb. Math. 196 (2005), no. 9-10, 1371–1402.
[CK00] N. G. Chebochko and M. I. Kuznetsov: Deformations of classical Lie algebras
(Russian). Mat. Sb. 191 (2000), no. 8, 69–88. English translation: Sb. Math. 191
(2000), no. 7-8, 1171–1190.
[CKK00] N. G. Chebochko, S. A. Kirillov and M. I. Kuznetsov: Deformations of a Lie algebra
of type G2 of characteristic three (Russian). Izv. Vyssh. Uchebn. Zaved. Mat. 2000,
no. 3, 33–38. English translation: Russian Math. (Iz. VUZ) 44 (2000), no. 3, 31–36.
[DG70] M. Demazure and P. Gabriel: Groupes algebriques. Tome I: Geometrie algebrique,
generalites, groupes commutatifs (French). Avec un appendice Corps de classes local
par Michiel Hazewinkel. Masson and Cie, Editeur, Paris; North-Holland Publishing
Co., Amsterdam, 1970.
[DK78] A. S. Dzumadildaev and A. I. Kostrikin: Deformations of the Lie algebra W1(m)
(Russian). Algebra, number theory and their applications. Trudy Mat. Inst. Steklov.
148 (1978), 141–155, 275.
[DZU80] A. S. Dzumadildaev: Deformations of general Lie algebras of Cartan type (Russian).
Dokl. Akad. Nauk SSSR 251 (1980), no. 6, 1289–1292. English translation: Soviet
Math. Dokl. 21 (1980), no. 2, 605–609.
[DZU81] A. S. Dzumadildaev: Relative cohomology and deformations of the Lie algebras of
Cartan types (Russian). Dokl. Akad. Nauk SSSR 257 (1981), no. 5, 1044–1048.
English translation: Soviet Math. Dokl. 23 (1981), no. 2, 398–402.
207
[DZU89] A. S: Dzumadildaev: Deformations of the Lie algebras Wn(m) (Russian). Mat. Sb.
180 (1989), no. 2, 168–186. English translation: Math. USSR-Sb. 66 (1990), no. 1,
169–187.
[FAR91] R. Farnsteiner: Cohomology groups of reduced enveloping algebras. Math. Zeit. 206
(1991), 103–117.
[FS88] R. Farnsteiner and H. Strade: Modular Lie algebras and their representation. Mono-
graphs and textbooks in pure and applied mathematics, vol. 116. Dekker, New York,
1988.
[FRA54] M. S. Frank: A new class of simple Lie algebras. Proc. Nat. Acad. Sci. USA 40
(1954), 713–719.
[FRA64] M. S. Frank: Two new classes of simple Lie algebras. Trans. Amer. Math. Soc. 112
(1964), 456–482.
[GER64] M. Gerstenhaber: On the deformation of rings and algebras. Ann. of Math. 79
(1964), 59–103.
[HS97] P. J. Hilton and U. A. Stammbach: A course in homological algebra. Second edition.
Graduate Texts in Mathematics, 4. Springer-Verlag, New York, 1997.
[HOC54] G. Hochschild: Cohomology of restricted Lie algebras. Amer. J. Math. 76 (1954),
555–580.
[JAC37] N. Jacobson: Abstract derivations and Lie algebras. Trans. Amer. Math. Soc. 42
(1937), 206–224.
[JAC80] N. Jacobson: Basic algebra II. W. H. Freeman and Co., San Francisco, Calif., 1980.
[JAC43] N. Jacobson: Classes of restricted Lie algebras of characteristic p, II. Duke Math.
Journal 10 (1943), 107-121.
[JAC71] N. Jacobson: Exceptional Lie algebras. Lecture Notes in Pure and Applied Mathe-
matics 1, Marcel Dekker, New York 1971.
[JAC62] N. Jacobson: Lie algebras. Interscience Tracts in Pure and Applied Mathematics
No. 10, New York-London, 1962.
[KAC70] V. G. Kac: The classification of the simple Lie algebras over a field with nonzero
characteristic (Russian). Izv. Akad. Nauk SSSR Ser. Mat. 34 (1970), 385–408.
English translation: Math. USSR-Izv. 4 (1970), 391–413.
208
[KAC74] V. G. Kac: Description of the filtered Lie algebras with which graded Lie algebras of
Cartan type are associated (Russian). Izv. Akad. Nauk SSSR Ser. Mat. 38 (1974),
800–834. English translation: Math. USSR-Izv. 8 (1974), 801–835.
[KAP71] I. Kaplansky: Lie algebras and locally compact groups. Univ. of Chicago Press,
Chicago, 1971.
[KS66] A. I. Kostrikin and I. R. Shafarevich: Cartan’s pseudogroups and the p-algebras of
Lie (Russian). Dokl. Akad. Nauk SSSR 168 (1966), 740–742. English translation:
Soviet Math. Dokl. 7 (1966), 715–718.
[KS69] A. I. Kostrikin and I. R. Shafarevich: Graded Lie algebras of finite characteristic
(Russian). Izv. Akad. Nauk SSSR Ser. Math. 33 (1969), 251–322. English transla-
tion: Math. USSR-Izv. 3 (1969), 237–304.
[MEL80] G. M. Melikian: Simple Lie algebras of characteristic 5 (Russian). Uspekhi Mat.
Nauk 35 (1980), no. 1 (211), 203–204.
[PS97] A. Premet and H. Strade: Simple Lie algebras of small characteristic. I. Sandwich
elements. J. Algebra 189 (1997), 419–480.
[PS99] A. Premet and H. Strade: Simple Lie algebras of small characteristic. II. Excep-
tional roots. J. Algebra 216 (1999), 190–301.
[PS01] A. Premet and H. Strade:Simple Lie algebras of small characteristic. III. The toral
rank 2 case. J. Algebra 242 (2001), 236–337.
[PS04] A. Premet and H. Strade: Simple Lie algebras of small characteristic. IV. Solvable
and classical roots. J. Algebra 278 (2004), 766–833.
[RUD71] A. N. Rudakov: Deformations of simple Lie algebras (Russian). Izv. Akad. Nauk
SSSR Ser. Mat. 35 (1971), 1113–1119.
[SEL67] G. B. Seligman: Modular Lie algebras. Ergebnisse der Mathematik und ihrer Gren-
zgebiete, Band 40. Springer-Verlag, New York, 1967.
[STR89] H. Strade: The Classification of the Simple Modular Lie Algebras: I. Determination
of the two-sections. Ann. of Math. 130 (1989), 643–677.
[STR92] H. Strade: The Classification of the Simple Modular Lie Algebras: II. The Toral
Structure. J. Algebra 151 (1992), 425–475.
[STR91] H. Strade: The Classification of the Simple Modular Lie Algebras: III. Solution of
the Classical Case. Ann. of Math. 133 (1991), 577–604.
209
[STR93] H. Strade: The Classification of the Simple Modular Lie Algebras: IV. Determining
the Associated Graded Algebra. Ann. of Math. 138 (1993), 1–59.
[STR94] H. Strade: The Classification of the Simple Modular Lie Algebras: V. Algebras with
Hamiltonian Two-sections. Abh. Math. Sem. Univ. Hamburg 64 (1994), 167–202.
[STR98] H. Strade: The classification of the simple modular Lie algebras. VI. Solving the
final case. Trans. Amer. Math. Soc. 350 (1998), 2553–2628.
[STR04] H. Strade: Simple Lie algebras over fields of positive characteristic I: Structure
theory. De Gruyter Expositions in Mathematics, 38. Walter de Gruyter, Berlin,
2004.
[SW91] H. Strade and R: L. Wilson: Classification of Simple Lie Algebras over Algebraically
Closed Fields of Prime Characteristic. Bull. Amer. Math. Soc. 24 (1991), 357–362.
[VIV1] F. Vivani: Deformations of restricted simple Lie algebras I. math.RA/0612861.
[VIV2] F. Viviani: Deformations of restricted simple Lie algebras II. math.RA/0702499.
[VIV3] F. Viviani: Deformations of the restricted Melikian algebra. math.RA/0702594.
[WEI78] B: Ju. Weisfailer: On the structure of the minimal ideal of some graded Lie algebras
of characteristic p > 0. J. Algebra 53 (1978), 344–361.
[WIL76] R. L. Wilson: A structural characterization of the simple Lie algebras of generalized
Cartan type over fields of prime characteristic. J. Algebra 40 (1976), 418–465.
210